soul
FlaggedEmbody this digital identity. Read SOUL.md first, then STYLE.md, then examples/. Become the person—opinions, voice, worldview.
examples/karpathy/data/writing/software-2.0.html
View on GitHub<!DOCTYPE html><html lang="en"><head><meta charSet="utf-8"/><meta charSet="utf-8"/><meta name="viewport" content="width=device-width, initial-scale=1"/><meta name="viewport" content="width=device-width,minimum-scale=1,initial-scale=1,maximum-scale=1"/><style data-precedence="high" data-href="lite-color-scheme-tokens">:root{color-scheme:light;--color-bg-neutral-primary: #ffffff;--color-bg-neutral-secondary: #f2f2f2;--color-bg-neutral-tertiary: #f9f9f9;--color-bg-neutral-quaternary: #242424;--color-bg-accent-secondary: #bbdbba;--color-bg-accent-tertiary: #d2e7d1;--color-bg-accent-quaternary: #e8f3e8;--color-bg-error-primary: #c94a4a;--color-bg-utility-yellow-primary: #fffae1;--color-bg-utility-blue-primary: #e5f2ff;--color-fg-neutral-primary: #242424;--color-fg-neutral-secondary: #6b6b6b;--color-fg-neutral-primary-hover: #000000;--color-fg-neutral-secondary-hover: #3f3f3f;--color-fg-neutral-tertiary: #ffffff;--color-fg-accent-primary: #1a8917;--color-fg-error-primary: #c94a4a;--color-border-neutral-primary: #f2f2f2;--color-border-neutral-secondary: #242424;--color-border-neutral-tertiary: #e5e5e5;--color-border-error-primary: #c94a4a}</style><script src="https://www.google.com/recaptcha/enterprise.js?render=6Le-uGgpAAAAAPprRaokM8AKthQ9KNGdoxaGUvVp" async=""></script><link rel="icon" href="https://miro.medium.com/v2/5d8de952517e8160e40ef9841c781cdc14a5db313057fa3c3de41c6f5b494b19"/><title>Medium</title><meta name="theme-color" content="#000000"/><meta name="twitter:app:name:iphone" content="Medium"/><meta name="twitter:app:id:iphone" content="828256236"/><meta property="al:ios:app_name" content="Medium"/><meta property="al:ios:app_store_id" content="828256236"/><meta property="al:android:package" content="com.medium.reader"/><meta property="al:android:app_name" content="Medium"/><meta property="fb:app_id" content="542599432471018"/><meta property="og:site_name" content="Medium"/><meta name="apple-itunes-app" content="app-id=828256236, app-argument=/software-2-0-a64152b37c35, affiliate-data=pt=698524&ct=smart_app_banner&mt=8"/><link rel="search" type="application/opensearchdescription+xml" title="Medium" href="/osd.xml"/><link rel="apple-touch-icon" sizes="152x152" href="https://miro.medium.com/v2/resize:fill:304:304/10fd5c419ac61637245384e7099e131627900034828f4f386bdaa47a74eae156"/><link rel="apple-touch-icon" sizes="120x120" href="https://miro.medium.com/v2/resize:fill:240:240/10fd5c419ac61637245384e7099e131627900034828f4f386bdaa47a74eae156"/><link rel="apple-touch-icon" sizes="76x76" href="https://miro.medium.com/v2/resize:fill:152:152/10fd5c419ac61637245384e7099e131627900034828f4f386bdaa47a74eae156"/><link rel="apple-touch-icon" sizes="60x60" href="https://miro.medium.com/v2/resize:fill:120:120/10fd5c419ac61637245384e7099e131627900034828f4f386bdaa47a74eae156"/><link rel="mask-icon" href="https://miro.medium.com/v2/resize:fill:1000:1000/7*GAOKVe--MXbEJmV9230oOQ.png" color="#171717"/><link rel="preconnect" href="https://glyph.medium.com" crossorigin=""/><link rel="manifest" href="/manifest.json"/><link rel="preconnect" href="https://www.google.com"/><link rel="preconnect" href="https://www.gstatic.com" crossorigin=""/><link id="glyph_preload_link" rel="preload" as="style" type="text/css" href="https://glyph.medium.com/css/unbound.css"/><title>Software 2.0. I sometimes see people refer to neural… | by Andrej Karpathy | Medium</title><meta property="og:type" content="article"/><meta property="article:published_time" content="2021-03-13T18:18:43.264Z"/><meta name="title" content="Software 2.0. I sometimes see people refer to neural… | by Andrej Karpathy | Medium"/><meta property="og:title" content="Software 2.0"/><meta property="al:android:url" content="medium://p/a64152b37c35"/><meta property="al:ios:url" content="medium://p/a64152b37c35"/><meta property="al:android:app_name" content="Medium"/><meta name="description" content="Software 2.0 I sometimes see people refer to neural networks as just “another tool in your machine learning toolbox”. They have some pros and cons, they work here or there, and sometimes you can …"/><meta property="og:description" content="I sometimes see people refer to neural networks as just “another tool in your machine learning toolbox”. They have some pros and cons, they…"/><meta property="og:url" content="https://karpathy.medium.com/software-2-0-a64152b37c35"/><meta property="al:web:url" content="https://karpathy.medium.com/software-2-0-a64152b37c35"/><meta property="og:image" content="https://miro.medium.com/v2/resize:fit:1017/1*CHcu2L0NmAZwCpQgmS1ByA.jpeg"/><meta property="article:author" content="https://karpathy.medium.com"/><meta name="author" content="Andrej Karpathy"/><meta name="robots" content="index,follow,max-image-preview:large"/><meta name="referrer" content="unsafe-url"/><meta property="twitter:title" content="Software 2.0"/><meta name="twitter:site" content="@Medium"/><meta name="twitter:app:url:iphone" content="medium://p/a64152b37c35"/><meta property="twitter:description" content="I sometimes see people refer to neural networks as just “another tool in your machine learning toolbox”. They have some pros and cons, they…"/><meta name="twitter:image:src" content="https://miro.medium.com/v2/resize:fit:1017/1*CHcu2L0NmAZwCpQgmS1ByA.jpeg"/><meta name="twitter:card" content="summary_large_image"/><meta name="twitter:creator" content="@karpathy"/><meta name="twitter:label1" content="Reading time"/><meta name="twitter:data1" content="9 min read"/><link rel="author" href="https://karpathy.medium.com"/><link rel="canonical" href="https://karpathy.medium.com/software-2-0-a64152b37c35"/><link rel="alternate" href="android-app://com.medium.reader/https/medium.com/p/a64152b37c35"/><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/entry.client-DtJLrZR8VY.js"/><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/vendor-react-CpuriD7yOR.js"/><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/RoutingContext-N-dlYRSyd6.js"/><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/index.esm-dkQXnoW74C.js"/><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/vendor-apollo-B0FLDVoLJa.js"/><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/catalog-B5naEjnKWy.js"/><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/vendor-lodash-C0lcM7XZQI.js"/><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/useClientViewerOrVisitorId-CBF3XYp5m3.js"/><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/Logo-CAikMcA5xF.js"/><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/entityPageRegistry-BzOK2jlsWT.js"/><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/actions-BuaTAfK8Ws.js"/><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/performance-B_nPciEuBN.js"/><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/useragent-DM_y0Y6BQs.js"/><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/actions-DarRc50fcc.js"/><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/actions-DGO5CiB3v-.js"/><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/actions-BBSCsZ7iAN.js"/><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/alternateUserTopBannerActions-CFUxDhHrPM.js"/><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/googleOneTapActions-D-Kvzy3lkp.js"/><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/actions-1GHcCLwm6f.js"/><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/action-CLBCv6uGgP.js"/><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/vendor-fela-Cy19ySZi4y.js"/><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/color-DcO1XOeLw1.js"/><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/root-BMBCWO_EnW.js"/><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/Redirect-CNnLIdBBDa.js"/><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/useBaseCacheControl.generated.graphql-mg_C73LFNG.js"/><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/selectors-Civm4lBZJ3.js"/><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/NotFoundHandler-D2FdvSX1Qb.js"/><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/GoogleAnalytics-CpX84ZJVzm.js"/><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/WithClientFlag-D6H5iZ5feY.js"/><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/Layer-BoAybuFPli.js"/><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/visits-BnXWJEFetB.js"/><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/useClientResponsiveSize-DMj2FEayo4.js"/><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/UserIdStorage-CucM5frZ7R.js"/><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/ErrorScreen-DfI8g0yVkR.js"/><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/index-DHOhXfp0T0.js"/><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/SusiClickable-D3bZIDOhCt.js"/><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/SusiModalContainer-DTz4QdX9F_.js"/><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/CloseButton-C6-EVtDglw.js"/><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/useReduxSusi-BDnRTMQqz7.js"/><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/RememberMeCheckbox-C1ARm7l_lx.js"/><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/GoogleRecaptcha-kaX0womyEh.js"/><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/FacebookSusiButton-Cdj6Ecnpls.js"/><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/TextField-BswpTNUiLv.js"/><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/email-x6DqV6Od7s.js"/><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/CircularImage-ChC4HYcXtN.js"/><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/ZoomableImage-CMLNdmn9hp.js"/><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/LoadingSpinner-C-Fc1l7DXa.js"/><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/Dialog-AkVTR0-Mir.js"/><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/SusiModal.generated.graphql-C2CdXisXOb.js"/><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/Toast-BeYgQ7UP8I.js"/><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/VisibilityController-DfDM1mH55T.js"/><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/Popover-BEHTzJA9dJ.js"/><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/miroImage-DnVurRI6Te.js"/><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/glyph-DCVZQIrAg0.js"/><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/isPost-Jkp00EWxnS.js"/><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/usePreference-D22I6RuVgo.js"/><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/useExperimentExposureLogger-RvbVwpzFqo.js"/><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/useGateExposureLogger-CNFBfDLsZy.js"/><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/NotFoundScreen-Cbxs6enf11.js"/><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/LoadingScreen-DXn01VrTWl.js"/><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/index-9bMYTUUBe-.js"/><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/GlobalNavigationContext-CFsD4sSJxy.js"/><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/useIsomorphicLayoutEffect-mTFvRtpWc0.js"/><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/OpenInApp-DfYOBi2m2t.js"/><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/useGoogleTagEvent-BVvaY-rQcL.js"/><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/SettingsPage-1nQFOpO_S1.js"/><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/PageHeader-vZiraAAZX8.js"/><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/PageNavigationBar-BE30YyYyoY.js"/><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/ResizeObserver-R46HWnzBPn.js"/><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/HorizontalScroller-DX9VVDGKwb.js"/><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/BasicTooltip-BwWRxZW0H2.js"/><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/TooltipPopover-BxIyQ3Eu21.js"/><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/converters-CFjGMkrk4S.js"/><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/PageTitle-x-SPpU_qxv.js"/><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/MaxWidth-clK7fEc193.js"/><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/useClientNavigation-BXw3WK3zM9.js"/><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/FriendBadge-BK5Uzu_yjN.js"/><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/useToggle-w-u1N56vw9.js"/><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/pluralize-Dcs29zKoPV.js"/><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/useTimer-ahkFCywLqc.js"/><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/useShimmerAnimation-BcTM3YJxhP.js"/><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/MembershipBadge-C58bOtXo9-.js"/><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/BookVerificationBadge-BcsZNIdQVu.js"/><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/index-C5HGGZaA1H.js"/><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/helpers-DuUr-IXW9T.js"/><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/formatMembershipDate-CKGLHYfZsK.js"/><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/timeFormatUtils-g2m1dSWL0J.js"/><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/BaseDialog-C9Jh0Y4bdK.js"/><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/ContentShimmer-Bv6xn6myOY.js"/><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/CreditCardIcon-BWlhbufPaa.js"/><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/visa-C-R6AjgiLa.js"/><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/memberships-DNbl3TjKMI.js"/><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/format-ORhoBzAfcE.js"/><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/useCurrentMembershipType-CAr67orssE.js"/><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/index-C8Wk7uW_bo.js"/><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/PresentationTracker-CPzAv_34HT.js"/><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/IntersectionObserver-BIUJnjgxm4.js"/><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/lists-filled-24px-DOUw0bo_Gh.js"/><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/SearchInline-Chjfzh1JdW.js"/><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/tinykeys.module-DmIGHpV-B2.js"/><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/PopoverMenu-gHTDy3R03r.js"/><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/SquareImage-BUQhjK_Bb3.js"/><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/MiroImage-BKAA34AFMI.js"/><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/topic-Dae92C93Dd.js"/><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/HorizontalDivider-HaOMjvrYve.js"/><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/explore-B7IXIWKhqc.js"/><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/Date-BkxdzGCRUJ.js"/><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/constant-BbXslOR1r4.js"/><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/CollectionAvatar-BbsSaHGDVu.js"/><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/publications-cDhzOwdN10.js"/><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/users-DYlYbYHnyO.js"/><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/isUserVerifiedBookAuthor-rzeUseXpbO.js"/><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/Modal-DTdLYpDmnA.js"/><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/FollowingBadge-PGSMEByPHA.js"/><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/useUserViewerEdge-CfC4ZfuTBQ.js"/><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/TopNav-AHlAsf9hna.js"/><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/theme-D11jH8i4PC.js"/><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/chevron-down-BNelhiVh37.js"/><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/plus-16px-0xnU0BlbCH.js"/><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/error-24px-BWFAb7mHfq.js"/><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/events-DFdwDo_p8W.js"/><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/useViewerQuery-CG78_WM4_E.js"/><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/twitter-icon-Bzemimo3um.js"/><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/Checkbox-CEpZybmOzs.js"/><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/useServiceAuthUrl-CYlhWiPOAI.js"/><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/facebook-icon-CTFtEBVMAi.js"/><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/FloatingPortal-CIHz4Ddj70.js"/><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/DefaultToastMessage-ClK0cOtZW0.js"/><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/useSendVerificationEmail-D-T6JolvWw.js"/><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/responsesSidebarUtils-P1lCDAk5tr.js"/><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/useMutingActions-QBw4WbTfeu.js"/><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/useCollectionViewerEdge-DSA5r45PFQ.js"/><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/useTagViewerEdge-BPwelZwdZz.js"/><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/cache-DAeQfZIjhg.js"/><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/UndeletePostMutation-DqELsRmbR9.js"/><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/DeletePostMutation-D_-pLOIrAI.js"/><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/DeletePostMutation.generated.graphql-CxsaFDqjlm.js"/><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/users-BgK6Dc8vo4.js"/><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/useFollowCollection-CGhsA02QO9.js"/><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/useSendFollowEvent-BMGhU8TxT5.js"/><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/useFollowUser-BgNqlRDj47.js"/><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/dom-DDy6wUDMDa.js"/><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/MiddotContainer-FtEBnsxBEZ.js"/><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/FollowSubscribeButton-Dooa_hj_C6.js"/><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/chevron-down-24px-CBba64dKHK.js"/><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/RadioGroup-DZhrVcchj7.js"/><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/numbers-C1x1ytAIkn.js"/><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/ShowEntitySlug-DrZts1oV_Z.js"/><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/createEntityRoute-BJd2kwGj0Q.js"/><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/context-Dpzg9QPdZT.js"/><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/ShowPost-DF5Gjy3lZX.js"><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/index-WLw2T4y-3o.js"><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/MastodonVerificationLink-Bm8g5wcSIu.js"><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/PostResultError-B9LHWmJIK2.js"><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/RedirectToSignIn-DyMm28ap10.js"><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/ParagraphStyleGrouping-Cy7Ets2O42.js"><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/TextParagraph-26KnX3Wa8D.js"><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/uri-Ctxdmxa3sb.js"><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/Code-DtBYAgnqsJ.js"><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/NormalizePostBodyModel-C0JY3W_nC5.js"><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/useUserAgentMobileOrTablet-B8RxRuG_3X.js"><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/ResponsesSidebarContext-CCbJ-I7EOF.js"><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/linkedin-filled-Cf9TQvBWl9.js"><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/response-16px-kK7rVmegXI.js"><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/lock-19px-ByXN80R6Mr.js"><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/WithClientViewer-CXK-DKFa6J.js"><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/ThreadedResponseContext-Dbb_dWSVZb.js"><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/UserMention-DrUM9dTqU2.js"><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/ConfirmDialog-JET23eK1mh.js"><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/TimeAgo-CRelcpedBq.js"><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/index-XhBPgZ8c4p.js"><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/index-Bf0PxdG7u-.js"><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/index-GO4FbZYrk5.js"><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/ResponsiveMiroImage-DJsjH_YNyU.js"><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/EmbedCatalog-CwUe8jBWwt.js"><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/PreviewCatalogCovers-CcHwNSLhCQ.js"><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/ResponsiveMediumBreakpointsMiroImage-D6mZ-PmouD.js"><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/FooterActions-CtJOR-LsOo.js"><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/PostPublishedDialog-C-Dpi_R4BS.js"><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/usePostViewerEdge-Bohjkz_23N.js"><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/friend-link-7sPV92ycLr.js"><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/ShareButton-DTLLcENQyA.js"><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/link-DdXm79w2NX.js"><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/InfoTooltip-C4m5FDHZyB.js"><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/info-circle-BQQlJhDVFj.js"><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/CopyUrl-Yx7fW1IwMw.js"><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/FooterLinks--jnSuYg7gV.js"><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/InfiniteScrollTrigger-CIoj4vmsrm.js"><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/TopNavUpsellBanner-CV-PLHusud.js"><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/OfferPill-BoOcEJCVen.js"><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/useMembershipDiscountUpsell-fFIMRt7yNU.js"><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/useEnableMembershipOffers-Cv3oFYYUnp.js"><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/DismissFlag.generated.graphql-CdLcxB9MtC.js"><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/PostBodyInserts-DF8FEPiT5J.js"><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/HighlightCTA-BZuOKMbVYZ.js"><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/UpsellPresentationTracker-Brpwy-JQU-.js"><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/featured-story-CN7P-sqps0.js"><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/OverlappingAvatars-zXpma7EC5H.js"><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/CollectionTooltip-6vFveop3SG.js"><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/PublicationFollowButton-BEVNkPfkge.js"><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/LoadableCollectionTooltip-UGIKeYE-5k.js"><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/OverflowMenuButton-CPiwJYq0G5.js"><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/ResponseConstants-BVJcI0zcXP.js"><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/FeatureStoryMenuItem-BWlBWgsJbY.js"><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/UserName-Dz-gl4Hjot.js"><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/UserUnblock.generated.graphql-BFEtMtAmb5.js"><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/OverflowDotsButton-DZa_6xLKGp.js"><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/ResponsiveIconButton-DrX6fhlm_I.js"><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/overflow-dots-filled-24px-CVPuedJZc3.js"><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/useClapCount-CaEq6aQn0B.js"><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/viewerEdge-OSXZlBOwtU.js"><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/BookmarkButton-kzwGk-MBCE.js"><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/bookmark-add-Oqv6jogtwo.js"><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/HideResponsesConfirmDialog-BZOPQL-Eo_.js"><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/CreateCatalogDialog-w8n0Vji6Mm.js"><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/helpers-C8qj4i3SA_.js"><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/RedirectToEditor-BlLyOSHLbs.js"><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/MuteTagModal-BEXfUz33W2.js"><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/RedErrorIcon-Dd2XKFfEGR.js"><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/NewsletterV3EmailToSubscribersMenuItem-B_Kja1SNIz.js"><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/WithdrawOrRemoveFromPublication-DdraMJICxe.js"><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/viewerStaffRole-CsjjdXpMaN.js"><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/helpers-D6fosCZ0jU.js"><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/MultiVote-ChdZNJSCvx.js"><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/MultiVoteCountDetail-CBl8_4mj05.js"><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/PostSharePopover-DV0_7-7dKk.js"><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/posts-uf1kCxzomP.js"><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/CopyShareLink-BbniN41TgC.js"><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/ShareLinkPopover-Cl5j5y_cNb.js"><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/share-24px-PXRq3ZJMfY.js"><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/FriendLink-C-39joZ2Pc.js"><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/DismissablePopover-BCcgteVFJh.js"><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/unlisted-19px-Cq96861jX0.js"><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/useUpdateRepostNote-DWMnTNg6Ua.js"><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/ElevateInteractable-m5IZ-AcWYx.js"><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/PostPreviewContainer-DxXUXaL8YG.js"><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/PostPresentationTracker-ChM7dJRfJO.js"><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/getGridPostPreviewImage-AnB2BOyEKE.js"><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/usePostPublishedAt-CksgteB2yL.js"><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/ElapsedTimeDate-DWqn6BoDcF.js"><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/index-DRTYG7q2St.js"><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/repost-24px-ng_ndc3xPw.js"><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/UserDismissablePopover-BE24JA-Qzh.js"><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/useRepostCount-D-CTncvvTY.js"><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/PostResponsesIconButton-DrNrjYIvPQ.js"><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/BaseResponsesIconButton-D65J-79wsU.js"><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/response-24px-B606iw33ko.js"><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/HeaderButtonRow-CO0mYk8-aN.js"><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/TagPillWithPopover-jVLSxse3r6.js"><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/mutations-CjsMPTfZM2.js"><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/useTopicReporting-BX2L2R8bHh.js"><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/StyledTagPill-BT1h1MgEP-.js"><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/getTagFeaturesDisplay-CoMKkb8W6_.js"><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/TagFollowButton-DgXQQhjoCD.js"><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/IosAppDownloadButton-CDro2ONTqo.js"><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/InResponseToEntityPreview-BGok38SZWg.js"><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/response-filled-19px-uieWVUivL5.js"><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/CatalogsListItem-BWL0wZE3bR.js"><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/lock-filled-11px-DF3eakpIEZ.js"><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/InteractivePostBody-CLl0norQTZ.js"><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/PostBody-D5YsG81Yoe.js"><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/PostBodySection-DqfMHHelJx.js"><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/useEnableTagPillsWithPopover-CBbZsEkdNl.js"><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/TagPillWithLink-Cp1QbSlce0.js"><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/metaDescription-DkZ7aiFopY.js"><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/titleHelpers-Ppoy2QVU9b.js"><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/useManagementPopoverMenuLauncher-TWMyeq6VUP.js"><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/link-21px-DLGeDxRN1C.js"><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/EditorEventContext-xF695FOV5g.js"><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/SimpleResponse-BLdrgCI2Lb.js"><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/DeleteResponseConfirmationDialog-D462ticLzW.js"><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/arrow-right-19px-CuNjuNVc6W.js"><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/overflow-dots-filled-25px-NB9btETzhh.js"><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/SuspendedBannerLoader-CEydsOr24u.js"><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/index-Bwwef313Iq.js"><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/PublisherLink-DqrLebZURX.js"><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/PublisherAvatar-CsZp54tOQl.js"><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/PublicationAvatar-Bd4bNEz39a.js"><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/PublicationColorBar-6DBLrsbypo.js"><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/StreamPostPreview-BdmIIPpsTX.js"><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/highlighter-DPWvEkIqP4.js"><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/getPreviewScaleHelper-oG4BMkSG5B.js"><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/StreamPostPreviewImage-D3KOapdgoR.js"><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/index-COMsBubzWV.js"><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/PostPreviewByLineAuthor-BUQBG2R6Sq.js"><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/PostPreviewByLineCollection-BBHLg48-jU.js"><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/PostPreviewFooterMeta-DMFuZGz9hx.js"><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/response-filled-16px-DuNjBRG3AI.js"><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/SeeMoreButton-n-fr3rPybc.js"><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/GridPostPreviewImage-RRDsx1hYm5.js"><link rel="modulepreload" href="https://cdn-client.medium.com/lite/assets/index-Cg8ISU_RWe.js"><style type="text/css" data-fela-rehydration="579" data-fela-type="STATIC">html{box-sizing:border-box;-webkit-text-size-adjust:100%}*, *:before, *:after{box-sizing:inherit}body{margin:0;padding:0;text-rendering:optimizeLegibility;-webkit-font-smoothing:antialiased;color:rgba(0,0,0,0.8);position:relative;min-height:100vh}h1, h2, h3, h4, h5, h6, dl, dd, ol, ul, menu, figure, blockquote, p, pre, form{margin:0}menu, ol, ul{padding:0;list-style:none;list-style-image:none}main{display:block}a{color:inherit;text-decoration:none}a, button, input{-webkit-tap-highlight-color:transparent}img, svg{vertical-align:middle}button{background:transparent;overflow:visible}button, input, optgroup, select, textarea{margin:0}:root{--reach-tabs:1;--reach-menu-button:1}#speechify-root{font-family:Sohne, sans-serif}div[data-popper-reference-hidden="true"]{visibility:hidden;pointer-events:none}.grecaptcha-badge{visibility:hidden}
/*XCode style (c) Angel Garcia <angelgarcia.mail@gmail.com>*/.hljs {background: #fff;color: black;
}/* Gray DOCTYPE selectors like WebKit */
.xml .hljs-meta {color: #c0c0c0;
}.hljs-comment,
.hljs-quote {color: #007400;
}.hljs-tag,
.hljs-attribute,
.hljs-keyword,
.hljs-selector-tag,
.hljs-literal,
.hljs-name {color: #aa0d91;
}.hljs-variable,
.hljs-template-variable {color: #3F6E74;
}.hljs-code,
.hljs-string,
.hljs-meta .hljs-string {color: #c41a16;
}.hljs-regexp,
.hljs-link {color: #0E0EFF;
}.hljs-title,
.hljs-symbol,
.hljs-bullet,
.hljs-number {color: #1c00cf;
}.hljs-section,
.hljs-meta {color: #643820;
}.hljs-title.class_,
.hljs-class .hljs-title,
.hljs-type,
.hljs-built_in,
.hljs-params {color: #5c2699;
}.hljs-attr {color: #836C28;
}.hljs-subst {color: #000;
}.hljs-formula {background-color: #eee;font-style: italic;
}.hljs-addition {background-color: #baeeba;
}.hljs-deletion {background-color: #ffc8bd;
}.hljs-selector-id,
.hljs-selector-class {color: #9b703f;
}.hljs-doctag,
.hljs-strong {font-weight: bold;
}.hljs-emphasis {font-style: italic;
}
</style><style type="text/css" data-fela-rehydration="579" data-fela-type="KEYFRAME">@-webkit-keyframes k1{0%{opacity:0.8}50%{opacity:0.5}100%{opacity:0.8}}@-moz-keyframes k1{0%{opacity:0.8}50%{opacity:0.5}100%{opacity:0.8}}@keyframes k1{0%{opacity:0.8}50%{opacity:0.5}100%{opacity:0.8}}</style><style type="text/css" data-fela-rehydration="579" data-fela-type="RULE">.a{font-family:medium-content-sans-serif-font, -apple-system, BlinkMacSystemFont, "Segoe UI", Roboto, Oxygen, Ubuntu, Cantarell, "Open Sans", "Helvetica Neue", sans-serif}.b{font-weight:400}.c{background-color:var(--color-bg-neutral-primary)}.d{display:none}.e{display:block}.f{position:sticky}.g{top:0}.h{z-index:500}.i{padding:0 24px}.j{align-items:center}.k{border-bottom:solid 1px var(--color-border-neutral-primary)}.t{height:41px}.u{line-height:20px}.v{margin-left:8px}.w{color:var(--color-fg-neutral-secondary)}.x{fill:inherit}.y{font-size:13px}.z{font-family:sohne, "Helvetica Neue", Helvetica, Arial, sans-serif}.ab{letter-spacing:inherit}.ac{padding:0}.ae{margin:0}.af{cursor:pointer}.ag:disabled{cursor:not-allowed}.ah:disabled{color:var(--color-fg-neutral-secondary)}.ai:disabled{fill:var(--color-fg-neutral-secondary)}.al{height:100%}.am{display:flex}.be{color:var(--color-bg-neutral-primary)}.bf{fill:var(--color-bg-neutral-primary)}.bg{background:#1A8917}.bh{border-color:#1A8917}.bl:disabled{cursor:inherit !important}.bm:disabled{opacity:0.3}.bn:disabled:hover{background:#1A8917}.bo:disabled:hover{border-color:#1A8917}.bp{border-radius:99em}.bq{border-width:1px}.br{border-style:solid}.bs{box-sizing:border-box}.bt{display:inline-block}.bu{text-decoration:none}.bv{text-align:center}.bw{margin-left:16px}.bx{color:inherit}.by{font-size:inherit}.bz{border:inherit}.ca{font-family:inherit}.cb{font-weight:inherit}.cc{gap:24px}.cd{height:57px}.ce{flex:1 0 auto}.cf{border:none}.cg{width:auto}.ch path{fill:var(--color-fg-neutral-primary)}.ci{height:25px}.cj{margin-left:24px}.cl{border-radius:20px}.cm{width:240px}.cn{background:var(--color-bg-neutral-tertiary)}.co path{fill:var(--color-fg-neutral-secondary)}.cq{outline:none}.cr{font-size:14px}.cs{width:100%}.ct{padding:10px 20px 10px 0}.cu{background-color:transparent}.cv{color:var(--color-fg-neutral-primary)}.cw::placeholder{color:var(--color-fg-neutral-secondary)}.cx{margin-left:12px}.cy{margin-right:12px}.db{gap:4px}.dc{padding:6px 12px}.dd{white-space:nowrap}.de{background-color:var(--color-bg-neutral-quaternary)}.df{width:24px}.dg{height:24px}.dh{flex-shrink:0}.di path{fill:var(--color-fg-neutral-tertiary)}.dj{color:var(--color-fg-neutral-tertiary)}.dm{position:relative}.dn{fill:var(--color-fg-neutral-secondary)}.do{gap:8px}.dr{position:absolute}.ds{width:1px}.dt{height:1px}.du{margin:-1px}.dv{overflow:hidden}.dw{clip:rect(0, 0, 0, 0)}.dx{border-width:0}.dy{margin-right:32px}.dz{background:transparent}.ea svg{margin-left:4px}.eb svg{fill:var(--color-fg-neutral-secondary)}.ed{box-shadow:inset 0 0 0 1px rgba(0, 0, 0, 0.05)}.ee{border-radius:50%}.ef{height:32px}.eg{width:32px}.ei{background-color:var(--color-bg-neutral-secondary)}.ej{flex:0 0 auto}.el{flex:1 1 auto}.es{position:fixed}.et{top:50%}.eu{right:16px}.ev{transform:translateY(-50%)}.ex{flex-direction:column}.ey{align-items:flex-end}.ez{padding:12px 4px}.fa{border-radius:999px}.fb{background-color:color-mix(in srgb, var(--color-bg-neutral-primary) 38%, transparent)}.fc{backdrop-filter:blur(10px)}.fd{-webkit-backdrop-filter:blur(10px)}.fe{opacity:1}.ff{pointer-events:auto}.fg{transition:opacity 0.18s ease, visibility 0.18s ease}.fi{width:16px}.fj{height:2px}.fk{border-radius:1px}.fl{background-color:var(--color-fg-neutral-primary)}.fm{transition:opacity 0.15s ease, background-color 0.15s ease}.fn{opacity:0.15}.fo{max-width:300px}.fp{max-height:70vh}.fq{border-radius:8px}.fr{box-shadow:0px 0px 4px rgba(0, 0, 0, 0.05), 0px 2px 8px rgba(0, 0, 0, 0.15)}.fs{opacity:0}.ft{visibility:hidden}.fu{pointer-events:none}.fv{overflow-y:auto}.fw{overscroll-behavior:contain}.fx{min-height:0}.fy{padding:12px 40px 12px 16px}.fz{list-style-type:none}.ga{padding:2px 0}.gb{width:4px}.gc{height:4px}.gd{background-color:var(--color-fg-accent-primary)}.ge{flex:1}.gf{min-width:0}.gg{max-height:20px}.gh{text-overflow:ellipsis}.gi{display:-webkit-box}.gj{-webkit-line-clamp:1}.gk{-webkit-box-orient:vertical}.gl{word-break:break-all}.gn{color:var(--color-fg-accent-primary)}.gt{justify-content:center}.gu{margin:0 24px}.gy{background:rgba(255, 255, 255, 1)}.gz{border:1px solid var(--color-border-neutral-primary)}.ha{border-radius:4px}.hb{box-shadow:0 1px 4px var(--color-border-neutral-primary)}.hc{max-height:100vh}.hd{left:0}.he{top:calc(100vh + 100px)}.hf{bottom:calc(100vh + 100px)}.hg{width:10px}.hh{word-break:break-word}.hi{word-wrap:break-word}.hj:after{display:block}.hk:after{content:""}.hl:after{clear:both}.hm{line-height:1.23}.hn{letter-spacing:0}.ho{font-style:normal}.hp{font-weight:700}.ip{justify-content:space-between}.ja{gap:12px}.jb{align-items:baseline}.jc{width:36px}.jd{height:36px}.je{border:2px solid rgba(255, 255, 255, 1)}.jf{z-index:0}.jg{box-shadow:none}.jh{border:1px solid rgba(0, 0, 0, 0.05)}.ji{margin-bottom:2px}.jj{flex-wrap:nowrap}.jl{width:12px}.jm{flex-wrap:wrap}.jn{padding-left:8px}.jo{padding-right:8px}.kp> *{flex-shrink:0}.kq{overflow-x:scroll}.kr::-webkit-scrollbar{display:none}.ks{scrollbar-width:none}.kt{-ms-overflow-style:none}.ku{max-width:74px}.kv{flex-direction:row}.kw{z-index:2}.kx{margin-right:4px}.la{-webkit-user-select:none}.lb{border:0}.lc{fill:rgba(117, 117, 117, 1)}.lf{outline:0}.lg{user-select:none}.lh> svg{pointer-events:none}.lq{cursor:progress}.lr{margin-left:32px}.ls{margin-left:4px}.lt{margin-top:0px}.lu{padding:4px 0}.ma{display:inline-flex}.mb{align-items:flex-start}.mh{max-width:100%}.mi{padding:8px 2px}.mj svg{color:var(--color-fg-neutral-secondary)}.na{line-height:1.58}.nb{letter-spacing:-0.004em}.nc{font-family:source-serif-pro, Georgia, Cambria, "Times New Roman", Times, serif}.nx{margin-bottom:-0.46em}.ny{margin-left:auto}.nz{margin-right:auto}.oa{max-width:1017px}.og{clear:both}.oi{cursor:zoom-in}.oj{z-index:auto}.ol{height:auto}.om{background-color:rgba(255, 255, 255, 1)}.on{max-width:850px}.oo{max-width:1219px}.op{line-height:1.12}.oq{letter-spacing:-0.022em}.or{font-weight:600}.pm{margin-bottom:-0.28em}.ps{text-decoration:underline}.pt{max-width:596px}.pu{margin-bottom:26px}.pv{margin-top:6px}.pw{padding:8px 16px}.px{border-radius:100px}.py{box-shadow:inset 0 0 0 1px var(--color-border-neutral-tertiary)}.qa{border-top:none}.qb{margin-bottom:50px}.qc{height:52px}.qd{max-height:52px}.qe{box-sizing:content-box}.qf{position:static}.qg{z-index:1}.qh{max-width:155px}.qp{margin:0 32px}.qq{margin-right:20px}.qr{margin-bottom:64px}.re{height:48px}.rf{width:48px}.rh{height:64px}.ri{width:64px}.rj{align-self:flex-end}.rp{padding-right:4px}.rq{font-weight:500}.sd{margin-top:8px}.se{white-space:pre-wrap}.sf{margin:0 8px}.sg{margin-top:16px}.sh{animation:k1 1.2s ease-in-out infinite}.si{padding:32px 0}.sj{border-radius:100%}.sk{margin-top:4px}.sl{margin-bottom:8px}.sm{height:8px}.sn{width:120px}.so{width:80px}.sp{margin-top:12px}.sq{height:10px}.sr{width:90%}.sx{height:0px}.ta{height:20px}.tb{width:220px}.tc{width:180px}.td{margin:24px 0}.te{gap:16px}.tf{height:148px}.tg{margin-bottom:32px}.tm{align-items:stretch}.uu{flex-grow:0}.uv{height:200px}.uw{margin-bottom:16px}.ux{width:158px}.uy{height:16px}.uz{width:288px}.va{width:296px}.vb{margin-bottom:56px}.vc{width:224px}.vd{border-bottom:solid 1px var(--color-border-neutral-tertiary)}.ve{margin-top:72px}.vf{padding:24px 0}.vg{margin-bottom:0px}.vh{margin-right:16px}.aj:hover:not(:disabled){color:var(--color-fg-neutral-primary)}.ak:hover:not(:disabled){fill:var(--color-fg-neutral-primary)}.bi:hover{background:#156D12}.bj:hover{border-color:#156D12}.bk:hover{cursor:pointer}.dp:hover{color:var(--color-fg-neutral-primary)}.dq:hover{fill:var(--color-fg-neutral-primary)}.ec:hover svg{fill:var(--color-fg-neutral-primary)}.eh:hover{background-color:rgba(0, 0, 0, 0.1)}.jk:hover{text-decoration:underline}.le:hover{fill:rgba(8, 8, 8, 1)}.lv:hover{fill:var(--color-fg-neutral-primary-hover)}.lw:hover p{color:var(--color-fg-neutral-primary-hover)}.lx:hover{color:var(--color-fg-neutral-secondary-hover)}.ly:hover path{stroke:var(--color-fg-neutral-secondary-hover) !important}.lz:hover{color:var(--color-fg-neutral-primary-hover)}.mk:hover svg{color:var(--color-fg-neutral-primary-hover)}.pz:hover{background-color:var(--color-bg-neutral-secondary)}.rg:hover{background-color:none}.cp:focus-within path{fill:var(--color-fg-neutral-primary)}.ld:focus{fill:rgba(8, 8, 8, 1)}.ml:focus svg{color:var(--color-fg-neutral-primary-hover)}.ok:focus{transform:scale(1.01)}.li:active{border-style:none}</style><style type="text/css" data-fela-rehydration="579" data-fela-type="RULE" media="all and (min-width: 1080px)">.s{display:none}.av{font-size:14px}.aw{line-height:20px}.bc{font-size:13px}.bd{padding:5px 12px}.dl{display:flex}.ek{display:block}.gs{margin-bottom:50px}.gx{max-width:680px}.ik{font-size:42px}.il{margin-top:1.19em}.im{margin-bottom:32px}.in{line-height:52px}.io{letter-spacing:-0.011em}.iy{align-items:center}.iz{flex-direction:row}.kb{border-top:solid 1px var(--color-bg-neutral-secondary)}.kc{border-bottom:solid 1px var(--color-bg-neutral-secondary)}.kd{margin:32px 0 0}.ke{padding:3px 8px}.kn> *{margin-right:24px}.ko> :last-child{margin-right:0}.lp{margin-top:0px}.mg{margin:0}.nt{font-size:20px}.nu{margin-top:2.14em}.nv{line-height:32px}.nw{letter-spacing:-0.003em}.of{margin-top:56px}.pi{font-size:24px}.pj{margin-top:1.95em}.pk{line-height:30px}.pl{letter-spacing:-0.016em}.pr{margin-top:0.94em}.qo{display:inline-block}.qu{margin-bottom:0}.qv{margin-right:20px}.rk{max-width:500px}.sb{line-height:24px}.sc{letter-spacing:0}.sw{margin-bottom:52px}.tl{padding-top:72px}.tz{width:calc(100% + 32px)}.ua{margin-left:-16px}.ub{margin-right:-16px}.uq{padding-left:16px}.ur{padding-right:16px}.us{flex-basis:50%}.ut{max-width:50%}</style><style type="text/css" data-fela-rehydration="579" data-fela-type="RULE" media="all and (max-width: 1079.98px)">.lo{margin-top:0px}.ql{display:none}.qn{display:inline-block}</style><style type="text/css" data-fela-rehydration="579" data-fela-type="RULE" media="all and (max-width: 903.98px)">.er{display:none}.ln{margin-top:0px}.qk{display:inline-block}</style><style type="text/css" data-fela-rehydration="579" data-fela-type="RULE" media="all and (max-width: 727.98px)">.ll{margin-top:0px}.lm{margin-right:0px}.qj{display:inline-block}.qm{display:none}</style><style type="text/css" data-fela-rehydration="579" data-fela-type="RULE" media="all and (max-width: 551.98px)">.l{display:flex}.m{justify-content:space-between}.an{font-size:13px}.ao{line-height:20px}.ax{padding:0px 8px 1px}.ck{display:none}.em{min-width:100%}.go{margin-bottom:2px}.hq{font-size:32px}.hr{margin-top:1.01em}.hs{margin-bottom:24px}.ht{line-height:38px}.hu{letter-spacing:-0.014em}.iq{align-items:flex-start}.ir{flex-direction:column-reverse}.jp{margin:24px -24px 0}.jq{padding:0}.kf> *{margin-right:8px}.kg> :last-child{margin-right:24px}.ky{margin-left:0px}.lj{margin-top:0px}.lk{margin-right:0px}.mc{margin:0}.mm{border:1px solid var(--color-border-neutral-primary)}.mn{border-radius:99em}.mo{padding:0px 16px 0px 12px}.mp{height:38px}.mq{align-items:center}.ms svg{margin-right:8px}.nd{font-size:18px}.ne{margin-top:1.56em}.nf{line-height:28px}.ng{letter-spacing:-0.003em}.ob{margin-top:40px}.os{font-size:20px}.ot{margin-top:1.2em}.ou{line-height:24px}.ov{letter-spacing:0}.pn{margin-top:0.67em}.qi{display:inline-block}.qt{flex-direction:column}.rc{margin-bottom:20px}.rd{margin-right:0}.ro{max-width:100%}.rr{font-size:24px}.rs{line-height:30px}.rt{letter-spacing:-0.016em}.ss{margin-bottom:40px}.sy{margin-bottom:30px}.th{padding-top:48px}.tn{width:calc(100% + 24px)}.to{margin-left:-12px}.tp{margin-right:-12px}.uc{padding-left:12px}.ud{padding-right:12px}.ue{flex-basis:100%}.mr:hover{border-color:var(--color-border-neutral-tertiary)}</style><style type="text/css" data-fela-rehydration="579" data-fela-type="RULE" media="all and (min-width: 904px) and (max-width: 1079.98px)">.r{display:none}.at{font-size:14px}.au{line-height:20px}.ba{font-size:13px}.bb{padding:5px 12px}.dk{display:flex}.eq{min-width:100%}.gr{margin-bottom:50px}.gw{max-width:680px}.if{font-size:42px}.ig{margin-top:1.19em}.ih{margin-bottom:32px}.ii{line-height:52px}.ij{letter-spacing:-0.011em}.iw{align-items:center}.ix{flex-direction:row}.jx{border-top:solid 1px var(--color-bg-neutral-secondary)}.jy{border-bottom:solid 1px var(--color-bg-neutral-secondary)}.jz{margin:32px 0 0}.ka{padding:3px 8px}.kl> *{margin-right:24px}.km> :last-child{margin-right:0}.mf{margin:0}.np{font-size:20px}.nq{margin-top:2.14em}.nr{line-height:32px}.ns{letter-spacing:-0.003em}.oe{margin-top:56px}.pe{font-size:24px}.pf{margin-top:1.95em}.pg{line-height:30px}.ph{letter-spacing:-0.016em}.pq{margin-top:0.94em}.qw{margin-bottom:0}.qx{margin-right:20px}.rl{max-width:500px}.rz{line-height:24px}.sa{letter-spacing:0}.sv{margin-bottom:52px}.tk{padding-top:72px}.tw{width:calc(100% + 32px)}.tx{margin-left:-16px}.ty{margin-right:-16px}.um{padding-left:16px}.un{padding-right:16px}.uo{flex-basis:50%}.up{max-width:50%}</style><style type="text/css" data-fela-rehydration="579" data-fela-type="RULE" media="all and (min-width: 728px) and (max-width: 903.98px)">.p{display:flex}.q{justify-content:space-between}.ar{font-size:13px}.as{line-height:20px}.az{padding:0px 8px 1px}.da{display:none}.ep{min-width:100%}.gq{margin-bottom:50px}.gv{max-width:680px}.ia{font-size:42px}.ib{margin-top:1.19em}.ic{margin-bottom:32px}.id{line-height:52px}.ie{letter-spacing:-0.011em}.iu{align-items:center}.iv{flex-direction:row}.jt{border-top:solid 1px var(--color-bg-neutral-secondary)}.ju{border-bottom:solid 1px var(--color-bg-neutral-secondary)}.jv{margin:32px 0 0}.jw{padding:3px 8px}.kj> *{margin-right:24px}.kk> :last-child{margin-right:0}.me{margin:0}.nl{font-size:20px}.nm{margin-top:2.14em}.nn{line-height:32px}.no{letter-spacing:-0.003em}.od{margin-top:56px}.pa{font-size:24px}.pb{margin-top:1.95em}.pc{line-height:30px}.pd{letter-spacing:-0.016em}.pp{margin-top:0.94em}.qy{margin-bottom:0}.qz{margin-right:20px}.rm{max-width:500px}.rx{line-height:24px}.ry{letter-spacing:0}.su{margin-bottom:52px}.tj{padding-top:72px}.tt{width:calc(100% + 28px)}.tu{margin-left:-14px}.tv{margin-right:-14px}.ui{padding-left:14px}.uj{padding-right:14px}.uk{flex-basis:50%}.ul{max-width:50%}</style><style type="text/css" data-fela-rehydration="579" data-fela-type="RULE" media="all and (min-width: 552px) and (max-width: 727.98px)">.n{display:flex}.o{justify-content:space-between}.ap{font-size:13px}.aq{line-height:20px}.ay{padding:0px 8px 1px}.cz{display:none}.eo{min-width:100%}.gp{margin-bottom:2px}.hv{font-size:32px}.hw{margin-top:1.01em}.hx{margin-bottom:24px}.hy{line-height:38px}.hz{letter-spacing:-0.014em}.is{align-items:flex-start}.it{flex-direction:column-reverse}.jr{margin:24px 0 0}.js{padding:0}.kh> *{margin-right:8px}.ki> :last-child{margin-right:8px}.kz{margin-left:0px}.md{margin:0}.mt{border:1px solid var(--color-border-neutral-primary)}.mu{border-radius:99em}.mv{padding:0px 16px 0px 12px}.mw{height:38px}.mx{align-items:center}.mz svg{margin-right:8px}.nh{font-size:18px}.ni{margin-top:1.56em}.nj{line-height:28px}.nk{letter-spacing:-0.003em}.oc{margin-top:40px}.ow{font-size:20px}.ox{margin-top:1.2em}.oy{line-height:24px}.oz{letter-spacing:0}.po{margin-top:0.67em}.qs{flex-direction:column}.ra{margin-bottom:20px}.rb{margin-right:0}.rn{max-width:100%}.ru{font-size:24px}.rv{line-height:30px}.rw{letter-spacing:-0.016em}.st{margin-bottom:40px}.sz{margin-bottom:30px}.ti{padding-top:48px}.tq{width:calc(100% + 24px)}.tr{margin-left:-12px}.ts{margin-right:-12px}.uf{padding-left:12px}.ug{padding-right:12px}.uh{flex-basis:100%}.my:hover{border-color:var(--color-border-neutral-tertiary)}</style><style type="text/css" data-fela-rehydration="579" data-fela-type="RULE" media="print">.ew{display:none}</style><style type="text/css" data-fela-rehydration="579" data-fela-type="RULE" media="(prefers-reduced-motion: reduce)">.fh{transition:none}</style><style type="text/css" data-fela-rehydration="579" data-fela-type="RULE" media="(orientation: landscape) and (max-width: 903.98px)">.gm{max-height:none}</style><style type="text/css" data-fela-rehydration="579" data-fela-type="RULE" media="(prefers-reduced-motion: no-preference)">.oh{transition:transform 300ms cubic-bezier(0.2, 0, 0.2, 1)}</style></head><body><!--$--><!--/$--><div class="a b c"><a href="/sitemap/sitemap.xml" class="d">Sitemap</a><link id="glyph_link" rel="stylesheet" type="text/css" href="https://glyph.medium.com/css/unbound.css"/><script>document.domain = document.domain;</script><div class="e f g h c"><div class="i j k l m n o p q r s t u"><a class="w x y z ab b ac ae af ag ah ai aj ak l n p r s j al u" href="https://play.google.com/store/apps/details?id=com.medium.reader&referrer=utm_source%3DmobileNavBar&source=---top_nav_layout_nav-----------------------------------------" rel="noopener follow">Open in app<svg width="10" height="10" viewBox="0 0 10 10" fill="none" xmlns="http://www.w3.org/2000/svg" class="v"><path d="M0.984835 8.48484C0.838388 8.63128 0.838388 8.86872 0.984835 9.01516C1.13128 9.16161 1.36872 9.16161 1.51517 9.01516L0.984835 8.48484ZM8.75 1.25H9.125C9.125 1.04289 8.95711 0.875 8.75 0.875V1.25ZM8.375 6.5C8.375 6.70711 8.54289 6.875 8.75 6.875C8.95711 6.875 9.125 6.70711 9.125 6.5H8.375ZM3.5 0.875C3.29289 0.875 3.125 1.04289 3.125 1.25C3.125 1.45711 3.29289 1.625 3.5 1.625V0.875ZM1.51517 9.01516L9.01516 1.51517L8.48484 0.984835L0.984835 8.48484L1.51517 9.01516ZM8.375 1.25V6.5H9.125V1.25H8.375ZM3.5 1.625H8.75V0.875H3.5V1.625Z" fill="currentColor"></path></svg></a><div class="am j"><p class="z b an ao ap aq ar as at au av aw w"><span data-dd-action-name="Susi presentation tracker global_nav"><button class="z b an ao ax ap aq ay ar as az ba au bb bc aw bd be bf bg bh bi bj bk bl bm bn bo bp bq br bs bt bu bv" data-testid="headerSignUpButton">Sign up</button></span></p><div class="bw e"><p class="z b an ao ap aq ar as at au av aw w"><span data-dd-action-name="Susi presentation tracker global_nav"><a class="bx x by bz ca ab cb ac ae af ag ah ai aj ak" data-testid="headerSignInButton" href="https://medium.com/m/signin?operation=login&redirect=https%3A%2F%2Fkarpathy.medium.com%2Fsoftware-2-0-a64152b37c35&source=post_page---top_nav_layout_nav-----------------------global_nav------------------" rel="noopener follow">Sign in</a></span></p></div></div></div><div class="i j k am cc cd"><div class="am j ce"><a class="bx x by cf ca ab cb ac ae af ag ah ai aj ak am" aria-label="Homepage" data-testid="headerMediumLogo" href="https://medium.com/?source=---top_nav_layout_nav-----------------------------------------" rel="noopener follow"><svg width="719" height="160" viewBox="0 0 719 160" fill="none" xmlns="http://www.w3.org/2000/svg" role="img" aria-labelledby="wordmark-medium-desc" class="cg ch ci"><desc id="wordmark-medium-desc">Medium Logo</desc><path d="M174.104 9.73376L174.319 9.68654V8.02097H130.39L89.6001 103.89L48.8105 8.02097H1.47153V9.68654L1.68401 9.73376C9.70237 11.5436 13.7732 14.2426 13.7732 23.9764V137.93C13.7732 147.664 9.68661 150.363 1.66825 152.173L1.45581 152.22V153.891H33.5738V152.226L33.3614 152.178C25.343 150.369 21.2722 147.669 21.2722 137.936V30.5862L73.6709 153.891H76.6427L130.568 27.1475V140.75C129.881 148.438 125.847 150.812 118.586 152.451L118.371 152.501V154.153H174.319V152.501L174.104 152.451C166.835 150.812 162.704 148.438 162.017 140.75L161.98 23.9764H162.017C162.017 14.2426 166.088 11.5436 174.104 9.73376ZM199.659 85.2223C200.574 64.7475 207.927 49.9698 220.265 49.7154C224.071 49.7783 227.263 51.0269 229.744 53.4295C235.016 58.5469 237.495 69.2406 237.112 85.2223H199.659ZM199.106 90.9928H264.679V90.7174C264.493 75.061 259.958 62.8826 251.213 54.5206C243.654 47.2944 232.462 43.318 220.706 43.318H220.443C214.342 43.318 206.859 44.7974 201.534 47.478C195.473 50.2846 190.127 54.4813 185.679 59.9895C178.518 68.8629 174.18 80.8551 173.125 94.3318C173.091 94.7357 173.062 95.1397 173.033 95.5436C173.005 95.9475 172.986 96.3095 172.968 96.6951C172.931 97.4164 172.902 98.1403 172.887 98.8669C172.86 100.031 172.853 101.204 172.874 102.384C173.584 132.908 190.072 157.301 219.357 157.301C245.062 157.301 260.032 138.51 263.764 113.288L261.878 112.624C255.321 126.18 243.544 134.395 230.14 133.393C211.843 132.024 197.826 113.471 199.098 90.998M338.82 132.357C336.669 137.458 332.181 140.265 326.167 140.265C320.152 140.265 314.654 136.136 310.749 128.635C306.552 120.582 304.344 109.199 304.344 95.7141C304.344 67.6485 313.073 49.4951 326.584 49.4951C332.241 49.4951 336.695 52.3016 338.82 57.1987V132.357ZM382.319 152.365C374.3 150.468 370.23 147.643 370.23 137.414V1.30884L321.514 15.6616V17.419L321.813 17.3954C328.533 16.8524 333.091 17.781 335.738 20.2256C337.81 22.1403 338.82 25.078 338.82 29.2118V47.8715C334.017 44.8052 328.304 43.3128 321.372 43.3128C307.313 43.3128 294.463 49.2328 285.196 59.9843C275.536 71.1895 270.429 86.5023 270.429 104.262C270.426 135.982 286.041 157.301 309.28 157.301C322.875 157.301 333.813 149.852 338.82 137.288V154.153H382.531V152.407L382.319 152.365ZM424.1 19.819C424.1 9.91474 416.632 2.44458 406.725 2.44458C396.866 2.44458 389.152 10.0774 389.152 19.819C389.152 29.5606 396.873 37.1934 406.725 37.1934C416.632 37.1934 424.1 29.7233 424.1 19.819ZM435.599 152.365C427.58 150.468 423.51 147.643 423.51 137.414H423.475V43.6354L379.761 56.1862V57.8911L380.024 57.9147C389.482 58.7567 392.071 62.0144 392.071 73.0675V154.153H435.822V152.407L435.599 152.365ZM547.612 152.365C539.594 150.468 535.523 147.643 535.523 137.414V43.6354L493.902 55.7718V57.482L494.148 57.5082C501.881 58.3213 504.115 61.7652 504.115 72.8682V132.147C501.537 137.249 496.7 140.278 490.841 140.483C481.338 140.483 476.105 134.064 476.105 122.41V43.638L432.391 56.1888V57.8911L432.653 57.9147C442.112 58.7541 444.703 62.0118 444.703 73.0675V123.237C444.681 126.739 444.985 130.236 445.613 133.681L446.4 137.104C450.101 150.366 459.798 157.301 474.99 157.301C487.858 157.301 499.137 149.335 504.105 136.871V154.182H547.819V152.435L547.612 152.365ZM717.43 154.153V152.404L717.217 152.354C708.517 150.348 705.128 146.565 705.128 138.864V75.0741C705.128 55.1843 693.957 43.3128 675.245 43.3128C661.605 43.3128 650.104 51.1947 645.676 63.4728C642.159 50.4629 632.037 43.3128 617.07 43.3128C603.924 43.3128 593.621 50.2505 589.201 61.9698V43.6433L545.487 55.68V57.3954L545.75 57.419C555.095 58.2479 557.797 61.6 557.797 72.3698V154.153H598.584V152.407L598.369 152.354C591.428 150.723 589.188 147.748 589.188 140.115V66.998C591.024 62.7095 594.725 57.6288 602.041 57.6288C611.127 57.6288 615.733 63.9239 615.733 76.3252V154.153H656.53V152.407L656.315 152.354C649.375 150.723 647.135 147.748 647.135 140.115V75.0662C647.152 72.634 646.959 70.2048 646.557 67.8059C648.504 63.1449 652.417 57.6288 660.032 57.6288C669.246 57.6288 673.723 63.7429 673.723 76.3252V154.153H717.43Z" fill="#242424"></path></svg></a><div class="cj ck"><div class="am cf cl cm cn j co cp"><div class="bt" aria-describedby="searchResults" aria-labelledby="searchResults" aria-haspopup="listbox"></div><div class="cx cy am"><svg width="24" height="24" viewBox="0 0 24 24" fill="none" xmlns="http://www.w3.org/2000/svg"><path fill-rule="evenodd" clip-rule="evenodd" d="M4.09234 11.06C4.09234 10.1473 4.27211 9.24358 4.62138 8.40036C4.97065 7.55715 5.48258 6.79099 6.12795 6.14562C6.77332 5.50026 7.53948 4.98832 8.38269 4.63905C9.2259 4.28978 10.1297 4.11001 11.0423 4.11001C11.955 4.11001 12.8588 4.28978 13.702 4.63905C14.5452 4.98832 15.3114 5.50026 15.9567 6.14562C16.6021 6.79099 17.114 7.55715 17.4633 8.40036C17.8126 9.24358 17.9923 10.1473 17.9923 11.06C17.9923 12.9033 17.2601 14.671 15.9567 15.9744C14.6534 17.2778 12.8856 18.01 11.0423 18.01C9.19909 18.01 7.43133 17.2778 6.12795 15.9744C4.82457 14.671 4.09234 12.9033 4.09234 11.06ZM11.0423 3.01001C9.75141 3.01124 8.4797 3.32292 7.33449 3.91876C6.18929 4.51459 5.20418 5.37711 4.46226 6.43355C3.72034 7.48999 3.24337 8.70937 3.07159 9.98883C2.89981 11.2683 3.03825 12.5703 3.47525 13.785C3.91224 14.9997 4.63497 16.0915 5.58245 16.9683C6.52994 17.8451 7.6744 18.4812 8.9193 18.8229C10.1642 19.1646 11.473 19.2019 12.7353 18.9316C13.9977 18.6613 15.1765 18.0915 16.1723 17.27L19.9223 21.02C19.9727 21.078 20.0345 21.1251 20.1039 21.1582C20.1732 21.1913 20.2487 21.2098 20.3255 21.2125C20.4023 21.2152 20.4788 21.2021 20.5504 21.1739C20.6219 21.1458 20.6868 21.1032 20.7412 21.0488C20.7955 20.9945 20.8381 20.9295 20.8662 20.858C20.8944 20.7865 20.9075 20.71 20.9048 20.6332C20.9021 20.5563 20.8836 20.4809 20.8505 20.4116C20.8174 20.3422 20.7704 20.2804 20.7123 20.23L16.9823 16.5C18.0418 15.3483 18.7426 13.9128 18.999 12.3691C19.2554 10.8253 19.0564 9.24032 18.4261 7.80795C17.7959 6.37559 16.7618 5.15798 15.4504 4.30407C14.1391 3.45017 12.6072 2.997 11.0423 3.00001V3.01001Z" fill="currentColor"></path></svg></div><input role="combobox" aria-controls="searchResults" aria-expanded="false" aria-label="search" data-testid="headerSearchInput" tabindex="0" class="cf cq z cr u cs ct cu cv cw" placeholder="Search" value=""/></div></div></div><div class="ck cz da"><button type="button" class="am j db dc bp cf af dd de" aria-label="Get the Medium app"><svg width="24" height="24" viewBox="0 0 24 24" fill="none" xmlns="http://www.w3.org/2000/svg" class="df dg dh di"><g clip-path="url(#clip0_4116_10682)"><path d="M9.75 3.9375H5.25C4.9019 3.9375 4.56806 4.07578 4.32192 4.32192C4.07578 4.56806 3.9375 4.9019 3.9375 5.25V9.75C3.9375 9.92236 3.97145 10.093 4.03741 10.2523C4.10337 10.4115 4.20005 10.5562 4.32192 10.6781C4.4438 10.8 4.58849 10.8966 4.74773 10.9626C4.90697 11.0286 5.07764 11.0625 5.25 11.0625H9.75C9.92236 11.0625 10.093 11.0286 10.2523 10.9626C10.4115 10.8966 10.5562 10.8 10.6781 10.6781C10.8 10.5562 10.8966 10.4115 10.9626 10.2523C11.0286 10.093 11.0625 9.92236 11.0625 9.75V5.25C11.0625 5.07764 11.0286 4.90697 10.9626 4.74773C10.8966 4.58849 10.8 4.4438 10.6781 4.32192C10.5562 4.20005 10.4115 4.10337 10.2523 4.03741C10.093 3.97145 9.92236 3.9375 9.75 3.9375ZM9.9375 9.75C9.9375 9.79973 9.91775 9.84742 9.88258 9.88258C9.84742 9.91775 9.79973 9.9375 9.75 9.9375H5.25C5.20027 9.9375 5.15258 9.91775 5.11742 9.88258C5.08225 9.84742 5.0625 9.79973 5.0625 9.75V5.25C5.0625 5.20027 5.08225 5.15258 5.11742 5.11742C5.15258 5.08225 5.20027 5.0625 5.25 5.0625H9.75C9.79973 5.0625 9.84742 5.08225 9.88258 5.11742C9.91775 5.15258 9.9375 5.20027 9.9375 5.25V9.75ZM9.75 12.9375H5.25C4.9019 12.9375 4.56806 13.0758 4.32192 13.3219C4.07578 13.5681 3.9375 13.9019 3.9375 14.25V18.75C3.9375 19.0981 4.07578 19.4319 4.32192 19.6781C4.56806 19.9242 4.9019 20.0625 5.25 20.0625H9.75C9.92236 20.0625 10.093 20.0286 10.2523 19.9626C10.4115 19.8966 10.5562 19.8 10.6781 19.6781C10.8 19.5562 10.8966 19.4115 10.9626 19.2523C11.0286 19.093 11.0625 18.9224 11.0625 18.75V14.25C11.0625 14.0776 11.0286 13.907 10.9626 13.7477C10.8966 13.5885 10.8 13.4438 10.6781 13.3219C10.5562 13.2 10.4115 13.1034 10.2523 13.0374C10.093 12.9714 9.92236 12.9375 9.75 12.9375ZM9.9375 18.75C9.9375 18.7997 9.91775 18.8474 9.88258 18.8826C9.84742 18.9177 9.79973 18.9375 9.75 18.9375H5.25C5.20027 18.9375 5.15258 18.9177 5.11742 18.8826C5.08225 18.8474 5.0625 18.7997 5.0625 18.75V14.25C5.0625 14.2003 5.08225 14.1526 5.11742 14.1174C5.15258 14.0823 5.20027 14.0625 5.25 14.0625H9.75C9.79973 14.0625 9.84742 14.0823 9.88258 14.1174C9.91775 14.1526 9.9375 14.2003 9.9375 14.25V18.75ZM18.75 3.9375H14.25C13.9019 3.9375 13.5681 4.07578 13.3219 4.32192C13.0758 4.56806 12.9375 4.9019 12.9375 5.25V9.75C12.9375 9.92236 12.9714 10.093 13.0374 10.2523C13.1034 10.4115 13.2 10.5562 13.3219 10.6781C13.4438 10.8 13.5885 10.8966 13.7477 10.9626C13.907 11.0286 14.0776 11.0625 14.25 11.0625H18.75C18.9224 11.0625 19.093 11.0286 19.2523 10.9626C19.4115 10.8966 19.5562 10.8 19.6781 10.6781C19.8 10.5562 19.8966 10.4115 19.9626 10.2523C20.0286 10.093 20.0625 9.92236 20.0625 9.75V5.25C20.0625 4.9019 19.9242 4.56806 19.6781 4.32192C19.4319 4.07578 19.0981 3.9375 18.75 3.9375ZM18.9375 9.75C18.9375 9.79973 18.9177 9.84742 18.8826 9.88258C18.8474 9.91775 18.7997 9.9375 18.75 9.9375H14.25C14.2003 9.9375 14.1526 9.91775 14.1174 9.88258C14.0823 9.84742 14.0625 9.79973 14.0625 9.75V5.25C14.0625 5.20027 14.0823 5.15258 14.1174 5.11742C14.1526 5.08225 14.2003 5.0625 14.25 5.0625H18.75C18.7997 5.0625 18.8474 5.08225 18.8826 5.11742C18.9177 5.15258 18.9375 5.20027 18.9375 5.25V9.75ZM12.9375 16.5V13.5C12.9375 13.3508 12.9968 13.2077 13.1023 13.1023C13.2077 12.9968 13.3508 12.9375 13.5 12.9375C13.6492 12.9375 13.7923 12.9968 13.8977 13.1023C14.0032 13.2077 14.0625 13.3508 14.0625 13.5V16.5C14.0625 16.6492 14.0032 16.7923 13.8977 16.8977C13.7923 17.0032 13.6492 17.0625 13.5 17.0625C13.3508 17.0625 13.2077 17.0032 13.1023 16.8977C12.9968 16.7923 12.9375 16.6492 12.9375 16.5ZM20.0625 15C20.0625 15.1492 20.0032 15.2923 19.8977 15.3977C19.7923 15.5032 19.6492 15.5625 19.5 15.5625H17.0625V19.5C17.0625 19.6492 17.0032 19.7923 16.8977 19.8977C16.7923 20.0032 16.6492 20.0625 16.5 20.0625H13.5C13.3508 20.0625 13.2077 20.0032 13.1023 19.8977C12.9968 19.7923 12.9375 19.6492 12.9375 19.5C12.9375 19.3508 12.9968 19.2077 13.1023 19.1023C13.2077 18.9968 13.3508 18.9375 13.5 18.9375H15.9375V13.5C15.9375 13.3508 15.9968 13.2077 16.1023 13.1023C16.2077 12.9968 16.3508 12.9375 16.5 12.9375C16.6492 12.9375 16.7923 12.9968 16.8977 13.1023C17.0032 13.2077 17.0625 13.3508 17.0625 13.5V14.4375H19.5C19.6492 14.4375 19.7923 14.4968 19.8977 14.6023C20.0032 14.7077 20.0625 14.8508 20.0625 15ZM20.0625 18V19.5C20.0625 19.6492 20.0032 19.7923 19.8977 19.8977C19.7923 20.0032 19.6492 20.0625 19.5 20.0625C19.3508 20.0625 19.2077 20.0032 19.1023 19.8977C18.9968 19.7923 18.9375 19.6492 18.9375 19.5V18C18.9375 17.8508 18.9968 17.7077 19.1023 17.6023C19.2077 17.4968 19.3508 17.4375 19.5 17.4375C19.6492 17.4375 19.7923 17.4968 19.8977 17.6023C20.0032 17.7077 20.0625 17.8508 20.0625 18Z" fill="black"></path></g><defs><clipPath id="clip0_4116_10682"><rect width="24" height="24" fill="white"></rect></clipPath></defs></svg><span class="z b cr u dj">Get app</span></button></div><div class="ck cz p dk dl"><span data-dd-action-name="Susi presentation tracker new_post_topnav"><a class="bx x by bz ca ab cb ac ae af ag ah ai aj ak" data-testid="headerWriteButton" href="https://medium.com/m/signin?operation=register&redirect=https%3A%2F%2Fmedium.com%2Fnew-story&source=---top_nav_layout_nav-----------------------new_post_topnav------------------" rel="noopener follow"><div class="z b cr u w dm dn am j do dp dq"><svg width="24" height="24" viewBox="0 0 24 24" fill="none" xmlns="http://www.w3.org/2000/svg"><path d="M14 4C14.2761 4 14.5 3.77614 14.5 3.5C14.5 3.22386 14.2761 3 14 3V4ZM21 10C21 9.72386 20.7761 9.5 20.5 9.5C20.2239 9.5 20 9.72386 20 10H21ZM14 3H4V4H14V3ZM3 4V20H4V4H3ZM4 21H20V20H4V21ZM21 20V10H20V20H21ZM20 21C20.5523 21 21 20.5523 21 20H20V21ZM3 20C3 20.5523 3.44771 21 4 21V20H3ZM4 3C3.44772 3 3 3.44772 3 4H4H4V3Z" fill="currentColor"></path><path d="M17.5 4.5L9.04173 12.9583C9.01429 12.9857 8.99361 13.0192 8.98134 13.056L8.15811 15.5257C8.09297 15.7211 8.2789 15.907 8.47434 15.8419L10.944 15.0187C10.9808 15.0064 11.0143 14.9857 11.0417 14.9583L19.5 6.5M17.5 4.5L19.8232 2.17678C19.9209 2.07915 20.0791 2.07915 20.1768 2.17678L21.8232 3.82322C21.9209 3.92085 21.9209 4.07915 21.8232 4.17678L19.5 6.5M17.5 4.5L19.5 6.5" stroke="currentColor"></path></svg>Write</div></a></span></div><div class="cz da r s"><a class="bx x by bz ca ab cb ac ae af ag ah ai aj ak" data-testid="headerSearchButton" href="https://medium.com/search?source=---top_nav_layout_nav-----------------------------------------" rel="noopener follow"><div class="z b cr u w dm dn am j do dp dq"><svg width="24" height="24" viewBox="0 0 24 24" fill="none" xmlns="http://www.w3.org/2000/svg"><path fill-rule="evenodd" clip-rule="evenodd" d="M4.09234 11.06C4.09234 10.1473 4.27211 9.24358 4.62138 8.40036C4.97065 7.55715 5.48258 6.79099 6.12795 6.14562C6.77332 5.50026 7.53948 4.98832 8.38269 4.63905C9.2259 4.28978 10.1297 4.11001 11.0423 4.11001C11.955 4.11001 12.8588 4.28978 13.702 4.63905C14.5452 4.98832 15.3114 5.50026 15.9567 6.14562C16.6021 6.79099 17.114 7.55715 17.4633 8.40036C17.8126 9.24358 17.9923 10.1473 17.9923 11.06C17.9923 12.9033 17.2601 14.671 15.9567 15.9744C14.6534 17.2778 12.8856 18.01 11.0423 18.01C9.19909 18.01 7.43133 17.2778 6.12795 15.9744C4.82457 14.671 4.09234 12.9033 4.09234 11.06ZM11.0423 3.01001C9.75141 3.01124 8.4797 3.32292 7.33449 3.91876C6.18929 4.51459 5.20418 5.37711 4.46226 6.43355C3.72034 7.48999 3.24337 8.70937 3.07159 9.98883C2.89981 11.2683 3.03825 12.5703 3.47525 13.785C3.91224 14.9997 4.63497 16.0915 5.58245 16.9683C6.52994 17.8451 7.6744 18.4812 8.9193 18.8229C10.1642 19.1646 11.473 19.2019 12.7353 18.9316C13.9977 18.6613 15.1765 18.0915 16.1723 17.27L19.9223 21.02C19.9727 21.078 20.0345 21.1251 20.1039 21.1582C20.1732 21.1913 20.2487 21.2098 20.3255 21.2125C20.4023 21.2152 20.4788 21.2021 20.5504 21.1739C20.6219 21.1458 20.6868 21.1032 20.7412 21.0488C20.7955 20.9945 20.8381 20.9295 20.8662 20.858C20.8944 20.7865 20.9075 20.71 20.9048 20.6332C20.9021 20.5563 20.8836 20.4809 20.8505 20.4116C20.8174 20.3422 20.7704 20.2804 20.7123 20.23L16.9823 16.5C18.0418 15.3483 18.7426 13.9128 18.999 12.3691C19.2554 10.8253 19.0564 9.24032 18.4261 7.80795C17.7959 6.37559 16.7618 5.15798 15.4504 4.30407C14.1391 3.45017 12.6072 2.997 11.0423 3.00001V3.01001Z" fill="currentColor"></path></svg><span class="dr ds dt ac du dv dw dd dx speechify-ignore">Search</span></div></a></div><div class="dy ck cz da"><div class="am j"><p class="z b an ao ap aq ar as at au av aw w"><span data-dd-action-name="Susi presentation tracker global_nav"><button class="z b an ao ax ap aq ay ar as az ba au bb bc aw bd be bf bg bh bi bj bk bl bm bn bo bp bq br bs bt bu bv" data-testid="headerSignUpButton">Sign up</button></span></p><div class="bw e"><p class="z b an ao ap aq ar as at au av aw w"><span data-dd-action-name="Susi presentation tracker global_nav"><a class="bx x by bz ca ab cb ac ae af ag ah ai aj ak" data-testid="headerSignInButton" href="https://medium.com/m/signin?operation=login&redirect=https%3A%2F%2Fkarpathy.medium.com%2Fsoftware-2-0-a64152b37c35&source=post_page---top_nav_layout_nav-----------------------global_nav------------------" rel="noopener follow">Sign in</a></span></p></div></div></div><div class="e"><div><div class="bt"><div tabindex="-1" class="cq"><button class="cf dz ac am j af dm ea eb ec" aria-label="user options menu" data-testid="headerUserIcon"><div class="e dm"><img alt="Unknown user" class="e bs ee ef eg ei" src="https://miro.medium.com/v2/resize:fill:64:64/1*dmbNkD5D-u45r44go_cf0g.png" width="32" height="32" loading="lazy"/><div class="ed ee e ef eg dr g cf eh"></div></div></button></div></div></div></div></div></div><div class="am"><div class="ej ck cz da r ek" style="width:0px"></div><div class="el cs em eo ep eq" style="width:calc(100% - 0px)"><div class="e"><nav class="er" aria-label="Table of contents"><button type="button" aria-label="Table of contents" aria-haspopup="true" aria-expanded="false" aria-controls="_R_fjaar5_" class="es et eu ev h ew am ex ey do ae ez cf af fa fb fc fd fe ff fg fh"><span aria-hidden="true" class="e fi fj fk fe fl fm fh"></span><span aria-hidden="true" class="e fi fj fk fn de fm fh"></span><span aria-hidden="true" class="e fi fj fk fn de fm fh"></span><span aria-hidden="true" class="e fi fj fk fn de fm fh"></span></button><div id="_R_fjaar5_" class="es et eu ev h ew am ex fo fp dv fq c fr fs ft fu fg fh" data-testid="toc-popover"><div class="fv fw fx fy"><ol class="ae ac fz"><li class="ga"><a class="gn x cr bz z ab b ac ae af ag ah ai aj ak am j do u" title="Ongoing transition" aria-current="location" href="/?source=post_page-----a64152b37c35---------------------------------------#196c" rel="noopener follow"><span aria-hidden="true" class="dh gb gc ee gd"></span><span class="ge gf z b cr u dv gg gh gi gj gk gl gm bx">Ongoing transition</span></a></li><li class="ga"><a class="w x cr bz z ab b ac ae af ag ah ai aj ak am j do u" title="The benefits of Software 2.0" href="/?source=post_page-----a64152b37c35---------------------------------------#0fac" rel="noopener follow"><span aria-hidden="true" class="dh gb gc ee cu"></span><span class="ge gf z b cr u dv gg gh gi gj gk gl gm bx">The benefits of Software 2.0</span></a></li><li class="ga"><a class="w x cr bz z ab b ac ae af ag ah ai aj ak am j do u" title="The limitations of Software 2.0" href="/?source=post_page-----a64152b37c35---------------------------------------#375e" rel="noopener follow"><span aria-hidden="true" class="dh gb gc ee cu"></span><span class="ge gf z b cr u dv gg gh gi gj gk gl gm bx">The limitations of Software 2.0</span></a></li><li class="ga"><a class="w x cr bz z ab b ac ae af ag ah ai aj ak am j do u" title="Programming in the 2.0 stack" href="/?source=post_page-----a64152b37c35---------------------------------------#385e" rel="noopener follow"><span aria-hidden="true" class="dh gb gc ee cu"></span><span class="ge gf z b cr u dv gg gh gi gj gk gl gm bx">Programming in the 2.0 stack</span></a></li></ol></div></div></nav><script type="application/ld+json">{"@context":"https://schema.org","@id":"https://karpathy.medium.com/software-2-0-a64152b37c35","@type":"SocialMediaPosting","image":["https://miro.medium.com/1*CHcu2L0NmAZwCpQgmS1ByA.jpeg"],"url":"https://karpathy.medium.com/software-2-0-a64152b37c35","dateCreated":"2017-11-11T22:18:53Z","datePublished":"2017-11-11T22:18:53Z","dateModified":"2021-03-13T18:18:43Z","headline":"Software 2.0","name":"Software 2.0","description":"Software 2.0 I sometimes see people refer to neural networks as just “another tool in your machine learning toolbox”. They have some pros and cons, they work here or there, and sometimes you can …","identifier":"a64152b37c35","author":{"@context":"https://schema.org","@id":"https://medium.com/@karpathy","@type":"Person","identifier":"karpathy","name":"Andrej Karpathy","url":"https://medium.com/@karpathy"},"creator":{"@context":"https://schema.org","@id":"https://medium.com/@karpathy","@type":"Person","identifier":"karpathy","name":"Andrej Karpathy","url":"https://medium.com/@karpathy"},"publisher":{"@context":"https://schema.org","@type":"Organization","@id":"https://medium.com","name":"Medium","url":"https://medium.com","logo":{"@type":"ImageObject","width":500,"height":110,"url":"https://miro.medium.com/v2/resize:fit:500/7%2AV1_7XP4snlmqrc_0Njontw.png"}},"mainEntityOfPage":"https://karpathy.medium.com/software-2-0-a64152b37c35","isAccessibleForFree":true}</script><div class="go gp gq gr gs e"><div class="am gt"><div class="gf cs gu gv gw gx"></div></div><article><div class="e"><div class="e"><span class="e"></span><section><div><div class="dr hd he hf hg fu"></div><div class="hh hi hj hk hl"><div class="am gt"><div class="gf cs gu gv gw gx"><div><h1 id="62ab" class="pw-post-title hm hn ho z hp hq hr hs ht hu hv hw hx hy hz ia ib ic id ie if ig ih ii ij ik il im in io cv" data-testid="storyTitle">Software 2.0</h1><div><div class="speechify-ignore am ip"><div class="speechify-ignore cs e"><div class="am iq ir is it iu iv iw ix iy iz ja"><div class="am j ja"><div class="am jb"><div><div class="bt"><div tabindex="-1" class="cq"><a rel="noopener follow" href="/?source=post_page---byline--a64152b37c35---------------------------------------" data-discover="true"><div class="e jc jd ee je jf"><div class="e dm"><img alt="Andrej Karpathy" class="e bs ee ef eg ei" src="https://miro.medium.com/v2/resize:fill:64:64/0*8ldFdx9B6FhSkQmV.jpeg" width="32" height="32" loading="lazy" data-testid="authorPhoto"/><div class="jg ee e ef eg dr g jh eh"></div></div></div></a></div></div></div></div><span class="z b cr u cv"><div class="ji am j"><div class="am j jj"><div class="am j"><div><div class="bt"><div tabindex="-1" class="cq"><span class="z b cr u cv"><a class="bx x by bz ca ab cb ac ae af ag ah ai jk" data-testid="authorName" rel="noopener follow" href="/?source=post_page---byline--a64152b37c35---------------------------------------" data-discover="true">Andrej Karpathy</a></span></div></div></div></div><div class="jl bt"></div></div></div></span></div><div class="am j jm"><span class="z b cr u w"><div class="am ce"><span data-testid="storyReadTime">9 min read</span><div class="jn jo e" aria-hidden="true"><span class="e" aria-hidden="true"><span class="z b cr u w">·</span></span></div><span data-testid="storyPublishDate">Nov 11, 2017</span></div></span></div></div><div class="am ip jp jq jr js jt ju jv jw jx jy jz ka kb kc kd ke"><div class="ck cz p dk dl j"><div class="am j"><div class="ku e"><div class="am j kv kw"><div class="pw-multi-vote-icon dm kx ky kz la"><span data-dd-action-name="Susi presentation tracker clap_footer"><a class="bx x by bz ca ab cb ac ae af ag ah ai aj ak" data-testid="headerClapButton" href="https://medium.com/m/signin?actionUrl=https%3A%2F%2Fmedium.com%2F_%2Fvote%2Fp%2Fa64152b37c35&operation=register&redirect=https%3A%2F%2Fkarpathy.medium.com%2Fsoftware-2-0-a64152b37c35&user=Andrej+Karpathy&userId=ac9d9a35533e&source=---header_actions--a64152b37c35---------------------clap_footer------------------" rel="noopener follow"><div><div class="bt"><div tabindex="-1" class="cq"><div class="lb af lc ld le lf ac lg lh li la" role="presentation"><svg width="24" height="24" viewBox="0 0 24 24" xmlns="http://www.w3.org/2000/svg" aria-label="clap"><path fill-rule="evenodd" clip-rule="evenodd" d="M11.3702 0.827583L12 3.2822L12.6298 0.827583H11.3702Z"></path><path fill-rule="evenodd" clip-rule="evenodd" d="M13.9159 3.95255L15.4386 1.84138L14.2552 1.45158L13.9159 3.95255Z"></path><path fill-rule="evenodd" clip-rule="evenodd" d="M8.58869 1.84055L10.1114 3.95172L9.77379 1.45076L8.58869 1.84055Z"></path><path fill-rule="evenodd" clip-rule="evenodd" d="M18.523 18.9192C17.6623 19.7799 16.7727 20.1664 15.9029 20.25C16.0419 20.1341 16.1776 20.01 16.3101 19.8776C18.6976 17.4891 19.1702 14.9269 17.7087 12.2547L16.7975 10.6516L16.0072 8.9799C15.7465 8.42046 15.8127 7.99921 16.2099 7.69218C16.2866 7.63312 16.3749 7.59082 16.469 7.56801C16.5631 7.5452 16.6609 7.54239 16.7561 7.55977C17.0392 7.60611 17.3023 7.79149 17.4836 8.06046L19.8472 12.2174C20.8229 13.8411 20.9884 16.4538 18.523 18.9192ZM7.52441 18.4822L3.36992 14.3278C3.25591 14.2118 3.1786 14.0648 3.14765 13.9051C3.1167 13.7454 3.13348 13.5802 3.19591 13.43C3.25833 13.2798 3.36363 13.1513 3.49864 13.0606C3.63365 12.9699 3.79239 12.921 3.95503 12.92C4.16606 12.92 4.37875 13.0011 4.54013 13.1617L6.69765 15.3192C6.76738 15.3812 6.85815 15.4142 6.9514 15.4114C7.04465 15.4087 7.13333 15.3704 7.1993 15.3044C7.26526 15.2385 7.30353 15.1498 7.30627 15.0566C7.30902 14.9633 7.27603 14.8725 7.21406 14.8028L5.05655 12.6445L3.60827 11.1962C3.4532 11.0414 3.36595 10.8314 3.36572 10.6124C3.36549 10.3933 3.45229 10.1831 3.60703 10.028C3.76177 9.87297 3.97177 9.78572 4.19084 9.78549C4.40991 9.78526 4.6201 9.87206 4.77517 10.0268L8.21296 13.4662C8.2468 13.5003 8.28703 13.5273 8.33134 13.5457C8.37566 13.5641 8.42317 13.5736 8.47117 13.5736C8.51916 13.5736 8.56668 13.5641 8.61099 13.5457C8.6553 13.5273 8.69553 13.5003 8.72937 13.4662C8.7634 13.4324 8.79041 13.3922 8.80884 13.3479C8.82726 13.3036 8.83675 13.256 8.83675 13.208C8.83675 13.16 8.82726 13.1125 8.80884 13.0682C8.79041 13.0239 8.7634 12.9837 8.72937 12.9498L5.29323 9.51287L4.3233 8.54377C4.16876 8.38892 4.08197 8.17909 4.08197 7.96032C4.08197 7.74155 4.16876 7.53171 4.3233 7.37687C4.4802 7.22625 4.68926 7.14214 4.90675 7.14214C5.12424 7.14214 5.33331 7.22625 5.4902 7.37687L6.46013 8.34514L9.8971 11.7813C9.93077 11.8156 9.97094 11.8429 10.0153 11.8615C10.0596 11.8801 10.1072 11.8897 10.1553 11.8897C10.2034 11.8897 10.251 11.8801 10.2953 11.8615C10.3397 11.8429 10.3798 11.8156 10.4135 11.7813C10.4818 11.7127 10.5202 11.6199 10.5202 11.5231C10.5202 11.4263 10.4818 11.3334 10.4135 11.2649L6.97655 7.82956C6.89975 7.75312 6.83891 7.66217 6.79756 7.56201C6.75622 7.46185 6.7352 7.35447 6.73572 7.24611C6.73507 7.13769 6.75595 7.03021 6.79715 6.92991C6.83835 6.82962 6.89904 6.7385 6.97572 6.66183C7.13084 6.50719 7.34095 6.42035 7.55999 6.42035C7.77903 6.42035 7.98914 6.50719 8.14427 6.66183L13.9308 12.4491C13.9922 12.5076 14.0721 12.5429 14.1567 12.549C14.2413 12.5552 14.3254 12.5318 14.3947 12.4829C14.464 12.434 14.5142 12.3626 14.5367 12.2809C14.5593 12.1991 14.5528 12.112 14.5183 12.0345L13.4011 9.67094C13.1404 9.11149 13.2066 8.69025 13.6047 8.38239C13.6817 8.32396 13.77 8.28204 13.864 8.25926C13.958 8.23648 14.0556 8.23331 14.1509 8.24997C14.4339 8.29632 14.6963 8.4817 14.8783 8.75149L17.0714 12.6114C18.3732 14.9915 17.9545 17.2003 15.7936 19.3611C14.6383 20.5173 13.1917 20.9882 11.6036 20.7283C10.1859 20.4916 8.73765 19.6955 7.52524 18.4822H7.52441ZM10.7503 5.9708L12.8706 8.09108C12.4593 8.59342 12.4047 9.26128 12.7423 9.98128L12.9616 10.4464L9.43861 6.92335C9.37757 6.81002 9.34438 6.68378 9.34179 6.55508C9.34179 6.33411 9.42786 6.12721 9.58344 5.9708C9.74082 5.82112 9.9497 5.73766 10.1669 5.73766C10.3841 5.73766 10.593 5.82112 10.7503 5.9708ZM18.1051 7.67563C17.7948 7.21466 17.3586 6.91756 16.8753 6.83894C16.6812 6.80632 16.4825 6.81388 16.2915 6.86114C16.1005 6.9084 15.9212 6.99436 15.7647 7.1137C15.4535 7.3537 15.2599 7.65659 15.1754 7.99508C14.9266 7.75301 14.611 7.59102 14.2692 7.52997C13.9911 7.4859 13.7061 7.52267 13.4483 7.6359L11.2668 5.45439C10.9747 5.16339 10.5792 5.00001 10.1669 5.00001C9.7546 5.00001 9.35911 5.16339 9.06703 5.45439C8.87539 5.64887 8.73887 5.89078 8.67144 6.15535C8.5271 6.00929 8.35536 5.89315 8.16606 5.81358C7.97675 5.73401 7.77361 5.69257 7.56826 5.69165C7.36292 5.69073 7.15941 5.73034 6.9694 5.8082C6.77938 5.88607 6.60661 6.00066 6.46096 6.14542C6.25382 6.35175 6.10947 6.6126 6.04468 6.8977C5.42068 6.2737 4.39613 6.2737 3.80772 6.86128C3.51632 7.15315 3.35266 7.54872 3.35266 7.96114C3.35266 8.37357 3.51632 8.76914 3.80772 9.06101C3.56937 9.16032 3.30703 9.29935 3.09268 9.5137C2.80164 9.80592 2.63823 10.2015 2.63823 10.614C2.63823 11.0264 2.80164 11.422 3.09268 11.7142L3.6091 12.229C3.33829 12.291 3.08902 12.4243 2.88721 12.6152C2.6854 12.8061 2.53842 13.0476 2.46154 13.3146C2.38467 13.5815 2.38071 13.8642 2.45008 14.1332C2.51945 14.4022 2.65961 14.6477 2.85599 14.8442L7.00965 18.9987C8.32965 20.3178 9.91861 21.1876 11.4852 21.4475C11.806 21.5011 12.1307 21.5282 12.456 21.5286C13.3051 21.5286 14.1087 21.33 14.8494 20.9493C15.0803 20.9824 15.3087 21.0031 15.5346 21.0031C16.8008 21.0031 17.9917 20.4834 19.0403 19.4356C21.8028 16.6731 21.5917 13.7021 20.4786 11.85L18.1051 7.67645V7.67563Z"></path></svg></div></div></div></div></a></span></div><div class="pw-multi-vote-count e lj lk ll lm ln lo lp"><p class="z b y u w"><span class="lq">--</span></p></div></div></div><div class="lr dy e"><div><div class="bt"><div tabindex="-1" class="cq"><button class="af lb fe lu am j dn lv lw" aria-label="responses"><svg width="24" height="24" viewBox="0 0 24 24" xmlns="http://www.w3.org/2000/svg" class="lt"><path d="M18.0064 16.8029C19.5393 15.347 20.24 13.4782 20.24 11.4818C20.24 7.35659 16.7088 4 12.1914 4C7.67395 4 4 7.35659 4 11.4818C4 15.6077 7.67395 18.9643 12.1914 18.9643C13.0081 18.9643 13.8134 18.8527 14.5839 18.6374C14.815 18.8368 15.064 19.0281 15.3284 19.1956C16.3876 19.8892 17.5312 20.24 18.7266 20.24C18.9505 20.232 19.1273 20.1284 19.2132 19.953C19.2554 19.8711 19.2738 19.7795 19.2664 19.688C19.2589 19.5966 19.2259 19.5089 19.1711 19.4347C18.6763 18.765 18.326 18.0714 18.1313 17.3777C18.0754 17.1816 18.0337 16.9817 18.0064 16.7798V16.8029ZM14.884 17.8585L14.8167 17.6353L14.502 17.7309C13.7529 17.9559 12.9738 18.0698 12.1906 18.069C8.16787 18.069 4.89943 15.1142 4.89943 11.4818C4.89943 7.84932 8.16787 4.89376 12.1906 4.89376C16.2051 4.89376 19.3033 7.85171 19.3033 11.4874C19.3033 13.2813 18.695 14.9556 17.2757 16.2073L17.081 16.3747V16.6299C17.081 16.6857 17.0811 16.7814 17.0973 16.9249C17.1216 17.1561 17.1784 17.4032 17.2514 17.6584C17.4055 18.2165 17.6488 18.7746 17.9732 19.3167C17.2189 19.2051 16.489 18.9261 15.8077 18.4716C15.532 18.2962 15.094 18.0889 14.8669 17.8816L14.884 17.8585Z"></path></svg><p class="z b y u w"><span class="pw-responses-count ls lt">192</span></p></button></div></div></div></div><div class="am j w"><div><div class="bt"><div tabindex="-1" class="cq"><span data-dd-action-name="Susi presentation tracker repost_header"><a class="bx x by bz ca ab cb ac ae af ag ah ai aj ak" href="https://medium.com/m/signin?actionUrl=https%3A%2F%2Fmedium.com%2F_%2Frepost%2Fp%2Fa64152b37c35&operation=register&redirect=https%3A%2F%2Fkarpathy.medium.com%2Fsoftware-2-0-a64152b37c35&user=Andrej+Karpathy&userId=ac9d9a35533e&source=---header_actions--a64152b37c35---------------------repost_header------------------" rel="noopener follow"><span class="lb dz bx af fe lg am j lh li la ac lx ly" aria-disabled="false" aria-label="Repost"><div class="dg df e dm"><svg xmlns="http://www.w3.org/2000/svg" width="24" height="24" viewBox="0 0 24 24" fill="none"><path d="M3 8.72727L5.64706 6M5.64706 6L8.29412 8.72727M5.64706 6V14.1818V15.2475C5.64706 15.9633 5.92589 16.6509 6.42442 17.1646C6.94277 17.6986 7.6553 18 8.39955 18H12" stroke="currentColor" stroke-linecap="round"></path><path d="M21 15.2727L18.3529 18M18.3529 18L15.7059 15.2727M18.3529 18V9.81818V8.75249C18.3529 8.03671 18.0741 7.34908 17.5756 6.83545C17.0572 6.30139 16.3447 6 15.6004 6L12 6" stroke="currentColor" stroke-linecap="round"></path></svg></div></span></a></span></div></div></div><div class="ls e"></div></div></div></div><div class="am j kf kg kh ki kj kk kl km kn ko kp kq kr ks kt"><div class="fi cz da r s"></div><div class="ck cz"><div><div class="bt"><div tabindex="-1" class="cq"><span data-dd-action-name="Susi presentation tracker bookmark_footer"><a class="bx x by bz ca ab cb ac ae af ag ah ai aj ak" data-testid="headerBookmarkButton" href="https://medium.com/m/signin?actionUrl=https%3A%2F%2Fmedium.com%2F_%2Fbookmark%2Fp%2Fa64152b37c35&operation=register&redirect=https%3A%2F%2Fkarpathy.medium.com%2Fsoftware-2-0-a64152b37c35&source=---header_actions--a64152b37c35---------------------bookmark_footer------------------" rel="noopener follow"><span class=""><svg width="25" height="25" viewBox="0 0 25 25" fill="none" xmlns="http://www.w3.org/2000/svg" class="w lz" aria-label="Add to list bookmark button"><path d="M18 2.5C18 2.22386 18.2239 2 18.5 2C18.7762 2 19 2.22386 19 2.5V5H21.5C21.7761 5 22 5.22386 22 5.5C22 5.77614 21.7761 6 21.5 6H19V8.5C19 8.77614 18.7762 9 18.5 9C18.2239 9 18 8.77614 18 8.5V6H15.5C15.2239 6 15 5.77614 15 5.5C15 5.22386 15.2239 5 15.5 5H18V2.5Z" fill="currentColor"></path><path d="M7 7C7 6.44772 7.44772 6 8 6H11.5C11.7761 6 12 5.77614 12 5.5C12 5.22386 11.7761 5 11.5 5H8C6.89543 5 6 5.89543 6 7V21C6 21.1905 6.10823 21.3644 6.27912 21.4486C6.45002 21.5327 6.65387 21.5125 6.80486 21.3963L12.5 17L18.1951 21.3963C18.3461 21.5125 18.55 21.5327 18.7209 21.4486C18.8918 21.3644 19 21.1905 19 21V12.5C19 12.2239 18.7761 12 18.5 12C18.2239 12 18 12.2239 18 12.5V19.9846L12.8049 15.9729C12.6251 15.8346 12.3749 15.8346 12.1951 15.9729L7 19.9846V7Z" fill="currentColor"></path></svg></span></a></span></div></div></div></div><div class="bs ma mb"><div class="e ce"><div class="am gt"><div class="mc md me mf mg mh gf cs"><div class="am"><span data-dd-action-name="Susi presentation tracker post_audio_button"><a class="bx x by bz ca ab cb ac ae af ag ah ai aj ak" href="https://medium.com/m/signin?actionUrl=https%3A%2F%2Fmedium.com%2Fplans%3Fdimension%3Dpost_audio_button%26postId%3Da64152b37c35&operation=register&redirect=https%3A%2F%2Fkarpathy.medium.com%2Fsoftware-2-0-a64152b37c35&source=---header_actions--a64152b37c35---------------------post_audio_button------------------" rel="noopener follow"><div><div class="bt"><div tabindex="-1" class="cq"><button aria-label="Listen" data-testid="audioPlayButton" class="bx dn by bz ca ab cb mi ae af ag bm mj mk lw ml mm mn mo mp l mq mr ms mt mu mv mw n mx my mz"><svg width="24" height="24" viewBox="0 0 24 24" fill="none" xmlns="http://www.w3.org/2000/svg"><path fill-rule="evenodd" clip-rule="evenodd" d="M3 12C3 7.02944 7.02944 3 12 3C16.9706 3 21 7.02944 21 12C21 16.9706 16.9706 21 12 21C7.02944 21 3 16.9706 3 12ZM12 2C6.47715 2 2 6.47715 2 12C2 17.5228 6.47715 22 12 22C17.5228 22 22 17.5228 22 12C22 6.47715 17.5228 2 12 2ZM15.376 12.416L10.7773 15.4818C10.4451 15.7033 10 15.4651 10 15.0657V8.93426C10 8.53491 10.4451 8.29672 10.7774 8.51823L15.376 11.584C15.6728 11.7819 15.6728 12.2181 15.376 12.416Z" fill="currentColor"></path></svg><div class="da r s"><p class="z b cr u w">Listen</p></div></button></div></div></div></a></span></div></div></div></div></div><div class="bt" aria-describedby="postFooterSocialMenu" aria-labelledby="postFooterSocialMenu"><div><div class="bt"><div tabindex="-1" class="cq"><button aria-controls="postFooterSocialMenu" aria-expanded="false" aria-label="Share Post" data-testid="headerSocialShareButton" class="bx dn by bz ca ab cb mi ae af ag bm mj mk lw ml mm mn mo mp l mq mr ms mt mu mv mw n mx my mz"><svg width="24" height="24" viewBox="0 0 24 24" fill="none" xmlns="http://www.w3.org/2000/svg"><path fill-rule="evenodd" clip-rule="evenodd" d="M15.2176 4.93118C15.1883 4.98289 15.1484 5.0278 15.1004 5.06293L15.1118 5.06859C15.024 5.12481 14.9269 5.15261 14.8204 5.14293C14.7004 5.13293 14.6004 5.08293 14.5204 5.01293L12.5004 2.99293V10.0629C12.5004 10.3429 12.2704 10.5629 12.0004 10.5629C11.7304 10.5629 11.5004 10.3429 11.5004 10.0629V3.02293L9.50042 5.02293C9.42474 5.09456 9.32654 5.13771 9.22259 5.14501C9.11863 5.1523 9.01537 5.12329 8.93042 5.06293H8.91042C8.82042 4.99293 8.75042 4.88293 8.75042 4.76293C8.74294 4.70553 8.74801 4.64719 8.76528 4.59195C8.78254 4.5367 8.81159 4.48586 8.85042 4.44293L11.6504 1.64293C11.7439 1.55132 11.8695 1.5 12.0004 1.5C12.1313 1.5 12.257 1.55132 12.3504 1.64293L15.1504 4.43293C15.1931 4.47602 15.2258 4.52782 15.2466 4.58479C15.2673 4.64175 15.2754 4.70252 15.2704 4.76293C15.2649 4.82212 15.2469 4.87946 15.2176 4.93118ZM15.1118 5.06859L15.1204 5.07293V5.06293C15.1175 5.06485 15.1147 5.06674 15.1118 5.06859ZM16 7.06293H17.5C18.0304 7.06293 18.5391 7.27364 18.9142 7.64871C19.2893 8.02379 19.5 8.53249 19.5 9.06293V19.0629C19.5 19.5934 19.2893 20.1021 18.9142 20.4771C18.5391 20.8522 18.0304 21.0629 17.5 21.0629H6.5C5.4 21.0629 4.5 20.1629 4.5 19.0629V9.06293C4.5 8.53249 4.71071 8.02379 5.08579 7.64871C5.46086 7.27364 5.96957 7.06293 6.5 7.06293H8C8.13188 7.06481 8.25769 7.11872 8.35 7.21293C8.44421 7.30524 8.49812 7.43104 8.5 7.56293C8.49812 7.69481 8.44421 7.82061 8.35 7.91293C8.25769 8.00713 8.13188 8.06105 8 8.06293H6.4C5.9 8.06293 5.5 8.46293 5.5 8.96293V19.1629C5.5 19.4016 5.59482 19.6305 5.7636 19.7993C5.93239 19.9681 6.16131 20.0629 6.4 20.0629H17.6C18.1 20.0629 18.5 19.6629 18.5 19.1629V8.96293C18.5 8.46293 18.1 8.06293 17.6 8.06293H16C15.8674 8.06293 15.7402 8.01025 15.6464 7.91648C15.5527 7.82271 15.5 7.69554 15.5 7.56293C15.5 7.43032 15.5527 7.30314 15.6464 7.20937C15.7402 7.11561 15.8674 7.06293 16 7.06293Z" fill="currentColor"></path></svg><div class="da r s"><p class="z b cr u w">Share</p></div></button></div></div></div></div></div></div></div></div></div></div><p id="de15" class="pw-post-body-paragraph na nb ho nc b nd ne nf ng nh ni nj nk nl nm nn no np nq nr ns nt nu nv nw nx hh cv">I sometimes see people refer to neural networks as just “another tool in your machine learning toolbox”. They have some pros and cons, they work here or there, and sometimes you can use them to win Kaggle competitions. Unfortunately, this interpretation completely misses the forest for the trees. Neural networks are not just another classifier, they represent the beginning of a fundamental shift in how we develop software. They are Software 2.0.</p><p id="2814" class="pw-post-body-paragraph na nb ho nc b nd ne nf ng nh ni nj nk nl nm nn no np nq nr ns nt nu nv nw nx hh cv">The “classical stack” of <strong class="nc hp">Software 1.0</strong> is what we’re all familiar with — it is written in languages such as Python, C++, etc. It consists of explicit instructions to the computer written by a programmer. By writing each line of code, the programmer identifies a specific point in program space with some desirable behavior.</p><figure class="ob oc od oe of og ny nz paragraph-image"><div role="button" tabindex="0" class="oh oi dm oj cs ok"><span class="dr ds dt ac du dv dw dd dx speechify-ignore">Press enter or click to view image in full size</span><div class="ny nz oa"><picture><source srcSet="https://miro.medium.com/v2/resize:fit:640/format:webp/1*CHcu2L0NmAZwCpQgmS1ByA.jpeg 640w, https://miro.medium.com/v2/resize:fit:720/format:webp/1*CHcu2L0NmAZwCpQgmS1ByA.jpeg 720w, https://miro.medium.com/v2/resize:fit:750/format:webp/1*CHcu2L0NmAZwCpQgmS1ByA.jpeg 750w, https://miro.medium.com/v2/resize:fit:786/format:webp/1*CHcu2L0NmAZwCpQgmS1ByA.jpeg 786w, https://miro.medium.com/v2/resize:fit:828/format:webp/1*CHcu2L0NmAZwCpQgmS1ByA.jpeg 828w, https://miro.medium.com/v2/resize:fit:1100/format:webp/1*CHcu2L0NmAZwCpQgmS1ByA.jpeg 1100w, https://miro.medium.com/v2/resize:fit:1400/format:webp/1*CHcu2L0NmAZwCpQgmS1ByA.jpeg 1400w" sizes="(min-resolution: 4dppx) and (max-width: 700px) 50vw, (-webkit-min-device-pixel-ratio: 4) and (max-width: 700px) 50vw, (min-resolution: 3dppx) and (max-width: 700px) 67vw, (-webkit-min-device-pixel-ratio: 3) and (max-width: 700px) 65vw, (min-resolution: 2.5dppx) and (max-width: 700px) 80vw, (-webkit-min-device-pixel-ratio: 2.5) and (max-width: 700px) 80vw, (min-resolution: 2dppx) and (max-width: 700px) 100vw, (-webkit-min-device-pixel-ratio: 2) and (max-width: 700px) 100vw, 700px" type="image/webp"/><source data-testid="og" srcSet="https://miro.medium.com/v2/resize:fit:640/1*CHcu2L0NmAZwCpQgmS1ByA.jpeg 640w, https://miro.medium.com/v2/resize:fit:720/1*CHcu2L0NmAZwCpQgmS1ByA.jpeg 720w, https://miro.medium.com/v2/resize:fit:750/1*CHcu2L0NmAZwCpQgmS1ByA.jpeg 750w, https://miro.medium.com/v2/resize:fit:786/1*CHcu2L0NmAZwCpQgmS1ByA.jpeg 786w, https://miro.medium.com/v2/resize:fit:828/1*CHcu2L0NmAZwCpQgmS1ByA.jpeg 828w, https://miro.medium.com/v2/resize:fit:1100/1*CHcu2L0NmAZwCpQgmS1ByA.jpeg 1100w, https://miro.medium.com/v2/resize:fit:1400/1*CHcu2L0NmAZwCpQgmS1ByA.jpeg 1400w" sizes="(min-resolution: 4dppx) and (max-width: 700px) 50vw, (-webkit-min-device-pixel-ratio: 4) and (max-width: 700px) 50vw, (min-resolution: 3dppx) and (max-width: 700px) 67vw, (-webkit-min-device-pixel-ratio: 3) and (max-width: 700px) 65vw, (min-resolution: 2.5dppx) and (max-width: 700px) 80vw, (-webkit-min-device-pixel-ratio: 2.5) and (max-width: 700px) 80vw, (min-resolution: 2dppx) and (max-width: 700px) 100vw, (-webkit-min-device-pixel-ratio: 2) and (max-width: 700px) 100vw, 700px"/><img alt="" class="cs mh ol om" width="700" height="162" loading="eager" role="presentation"/></picture></div></div></figure><p id="b05f" class="pw-post-body-paragraph na nb ho nc b nd ne nf ng nh ni nj nk nl nm nn no np nq nr ns nt nu nv nw nx hh cv">In contrast, <strong class="nc hp">Software 2.0</strong> is written in much more abstract, human unfriendly language, such as the weights of a neural network. No human is involved in writing this code because there are a lot of weights (typical networks might have millions), and coding directly in weights is kind of hard (I tried).</p><figure class="ob oc od oe of og ny nz paragraph-image"><div role="button" tabindex="0" class="oh oi dm oj cs ok"><span class="dr ds dt ac du dv dw dd dx speechify-ignore">Press enter or click to view image in full size</span><div class="ny nz on"><picture><source srcSet="https://miro.medium.com/v2/resize:fit:640/format:webp/1*6EB1Xue1wM_QP0IIzXphQA.png 640w, https://miro.medium.com/v2/resize:fit:720/format:webp/1*6EB1Xue1wM_QP0IIzXphQA.png 720w, https://miro.medium.com/v2/resize:fit:750/format:webp/1*6EB1Xue1wM_QP0IIzXphQA.png 750w, https://miro.medium.com/v2/resize:fit:786/format:webp/1*6EB1Xue1wM_QP0IIzXphQA.png 786w, https://miro.medium.com/v2/resize:fit:828/format:webp/1*6EB1Xue1wM_QP0IIzXphQA.png 828w, https://miro.medium.com/v2/resize:fit:1100/format:webp/1*6EB1Xue1wM_QP0IIzXphQA.png 1100w, https://miro.medium.com/v2/resize:fit:1400/format:webp/1*6EB1Xue1wM_QP0IIzXphQA.png 1400w" sizes="auto, (min-resolution: 4dppx) and (max-width: 700px) 50vw, (-webkit-min-device-pixel-ratio: 4) and (max-width: 700px) 50vw, (min-resolution: 3dppx) and (max-width: 700px) 67vw, (-webkit-min-device-pixel-ratio: 3) and (max-width: 700px) 65vw, (min-resolution: 2.5dppx) and (max-width: 700px) 80vw, (-webkit-min-device-pixel-ratio: 2.5) and (max-width: 700px) 80vw, (min-resolution: 2dppx) and (max-width: 700px) 100vw, (-webkit-min-device-pixel-ratio: 2) and (max-width: 700px) 100vw, 700px" type="image/webp"/><source data-testid="og" srcSet="https://miro.medium.com/v2/resize:fit:640/1*6EB1Xue1wM_QP0IIzXphQA.png 640w, https://miro.medium.com/v2/resize:fit:720/1*6EB1Xue1wM_QP0IIzXphQA.png 720w, https://miro.medium.com/v2/resize:fit:750/1*6EB1Xue1wM_QP0IIzXphQA.png 750w, https://miro.medium.com/v2/resize:fit:786/1*6EB1Xue1wM_QP0IIzXphQA.png 786w, https://miro.medium.com/v2/resize:fit:828/1*6EB1Xue1wM_QP0IIzXphQA.png 828w, https://miro.medium.com/v2/resize:fit:1100/1*6EB1Xue1wM_QP0IIzXphQA.png 1100w, https://miro.medium.com/v2/resize:fit:1400/1*6EB1Xue1wM_QP0IIzXphQA.png 1400w" sizes="auto, (min-resolution: 4dppx) and (max-width: 700px) 50vw, (-webkit-min-device-pixel-ratio: 4) and (max-width: 700px) 50vw, (min-resolution: 3dppx) and (max-width: 700px) 67vw, (-webkit-min-device-pixel-ratio: 3) and (max-width: 700px) 65vw, (min-resolution: 2.5dppx) and (max-width: 700px) 80vw, (-webkit-min-device-pixel-ratio: 2.5) and (max-width: 700px) 80vw, (min-resolution: 2dppx) and (max-width: 700px) 100vw, (-webkit-min-device-pixel-ratio: 2) and (max-width: 700px) 100vw, 700px"/><img alt="" class="cs mh ol om" width="700" height="435" loading="lazy" role="presentation"/></picture></div></div></figure><p id="eabb" class="pw-post-body-paragraph na nb ho nc b nd ne nf ng nh ni nj nk nl nm nn no np nq nr ns nt nu nv nw nx hh cv">Instead, our approach is to specify some goal on the behavior of a desirable program (e.g., “satisfy a dataset of input output pairs of examples”, or “win a game of Go”), write a rough skeleton of the code (i.e. a neural net architecture) that identifies a subset of program space to search, and use the computational resources at our disposal to search this space for a program that works. In the case of neural networks, we restrict the search to a continuous subset of the program space where the search process can be made (somewhat surprisingly) efficient with backpropagation and stochastic gradient descent.</p><figure class="ob oc od oe of og ny nz paragraph-image"><div role="button" tabindex="0" class="oh oi dm oj cs ok"><span class="dr ds dt ac du dv dw dd dx speechify-ignore">Press enter or click to view image in full size</span><div class="ny nz oo"><picture><source srcSet="https://miro.medium.com/v2/resize:fit:640/format:webp/1*5NG3U8MsaTqmQpjkr_-UOw.png 640w, https://miro.medium.com/v2/resize:fit:720/format:webp/1*5NG3U8MsaTqmQpjkr_-UOw.png 720w, https://miro.medium.com/v2/resize:fit:750/format:webp/1*5NG3U8MsaTqmQpjkr_-UOw.png 750w, https://miro.medium.com/v2/resize:fit:786/format:webp/1*5NG3U8MsaTqmQpjkr_-UOw.png 786w, https://miro.medium.com/v2/resize:fit:828/format:webp/1*5NG3U8MsaTqmQpjkr_-UOw.png 828w, https://miro.medium.com/v2/resize:fit:1100/format:webp/1*5NG3U8MsaTqmQpjkr_-UOw.png 1100w, https://miro.medium.com/v2/resize:fit:1400/format:webp/1*5NG3U8MsaTqmQpjkr_-UOw.png 1400w" sizes="auto, (min-resolution: 4dppx) and (max-width: 700px) 50vw, (-webkit-min-device-pixel-ratio: 4) and (max-width: 700px) 50vw, (min-resolution: 3dppx) and (max-width: 700px) 67vw, (-webkit-min-device-pixel-ratio: 3) and (max-width: 700px) 65vw, (min-resolution: 2.5dppx) and (max-width: 700px) 80vw, (-webkit-min-device-pixel-ratio: 2.5) and (max-width: 700px) 80vw, (min-resolution: 2dppx) and (max-width: 700px) 100vw, (-webkit-min-device-pixel-ratio: 2) and (max-width: 700px) 100vw, 700px" type="image/webp"/><source data-testid="og" srcSet="https://miro.medium.com/v2/resize:fit:640/1*5NG3U8MsaTqmQpjkr_-UOw.png 640w, https://miro.medium.com/v2/resize:fit:720/1*5NG3U8MsaTqmQpjkr_-UOw.png 720w, https://miro.medium.com/v2/resize:fit:750/1*5NG3U8MsaTqmQpjkr_-UOw.png 750w, https://miro.medium.com/v2/resize:fit:786/1*5NG3U8MsaTqmQpjkr_-UOw.png 786w, https://miro.medium.com/v2/resize:fit:828/1*5NG3U8MsaTqmQpjkr_-UOw.png 828w, https://miro.medium.com/v2/resize:fit:1100/1*5NG3U8MsaTqmQpjkr_-UOw.png 1100w, https://miro.medium.com/v2/resize:fit:1400/1*5NG3U8MsaTqmQpjkr_-UOw.png 1400w" sizes="auto, (min-resolution: 4dppx) and (max-width: 700px) 50vw, (-webkit-min-device-pixel-ratio: 4) and (max-width: 700px) 50vw, (min-resolution: 3dppx) and (max-width: 700px) 67vw, (-webkit-min-device-pixel-ratio: 3) and (max-width: 700px) 65vw, (min-resolution: 2.5dppx) and (max-width: 700px) 80vw, (-webkit-min-device-pixel-ratio: 2.5) and (max-width: 700px) 80vw, (min-resolution: 2dppx) and (max-width: 700px) 100vw, (-webkit-min-device-pixel-ratio: 2) and (max-width: 700px) 100vw, 700px"/><img alt="" class="cs mh ol om" width="700" height="402" loading="lazy" role="presentation"/></picture></div></div></figure><p id="26d1" class="pw-post-body-paragraph na nb ho nc b nd ne nf ng nh ni nj nk nl nm nn no np nq nr ns nt nu nv nw nx hh cv">To make the analogy explicit, in Software 1.0, human-engineered source code (e.g. some .cpp files) is compiled into a binary that does useful work. In Software 2.0 most often the source code comprises 1) the dataset that defines the desirable behavior and 2) the neural net architecture that gives the rough skeleton of the code, but with many details (the weights) to be filled in. The process of training the neural network compiles the dataset into the binary — the final neural network. In most practical applications today, the neural net architectures and the training systems are increasingly standardized into a commodity, so most of the active “software development” takes the form of curating, growing, massaging and cleaning labeled datasets. This is fundamentally altering the programming paradigm by which we iterate on our software, as the teams split in two: the 2.0 programmers (data labelers) edit and grow the datasets, while a few 1.0 programmers maintain and iterate on the surrounding training code infrastructure, analytics, visualizations and labeling interfaces.</p><p id="d3ca" class="pw-post-body-paragraph na nb ho nc b nd ne nf ng nh ni nj nk nl nm nn no np nq nr ns nt nu nv nw nx hh cv">It turns out that a large portion of real-world problems have the property that it is significantly easier to collect the data (or more generally, identify a desirable behavior) than to explicitly write the program. Because of this and many other benefits of Software 2.0 programs that I will go into below, we are witnessing a massive transition across the industry where of a lot of 1.0 code is being ported into 2.0 code. Software (1.0) is eating the world, and now AI (Software 2.0) is eating software.</p><h2 id="196c" class="op oq ho z or os ot ou ov ow ox oy oz pa pb pc pd pe pf pg ph pi pj pk pl pm cv">Ongoing transition</h2><p id="f0ad" class="pw-post-body-paragraph na nb ho nc b nd pn nf ng nh po nj nk nl pp nn no np pq nr ns nt pr nv nw nx hh cv">Let’s briefly examine some concrete examples of this ongoing transition. In each of these areas we’ve seen improvements over the last few years when we give up on trying to address a complex problem by writing explicit code and instead transition the code into the 2.0 stack.</p><p id="ebff" class="pw-post-body-paragraph na nb ho nc b nd ne nf ng nh ni nj nk nl nm nn no np nq nr ns nt nu nv nw nx hh cv"><strong class="nc hp">Visual Recognition</strong> used to consist of engineered features with a bit of machine learning sprinkled on top at the end (e.g., an SVM). Since then, we discovered much more powerful visual features by obtaining large datasets (e.g. ImageNet) and searching in the space of Convolutional Neural Network architectures. More recently, we don’t even trust ourselves to hand-code the architectures and we’ve begun <a class="bx ps" href="https://arxiv.org/abs/1703.01041" rel="noopener ugc nofollow" target="_blank">searching over those</a> as well.</p><p id="6785" class="pw-post-body-paragraph na nb ho nc b nd ne nf ng nh ni nj nk nl nm nn no np nq nr ns nt nu nv nw nx hh cv"><strong class="nc hp">Speech recognition</strong> used to involve a lot of preprocessing, gaussian mixture models and hidden markov models, but <a class="bx ps" href="https://github.com/syhw/wer_are_we" rel="noopener ugc nofollow" target="_blank">today </a>consist almost entirely of neural net stuff. A very related, often cited humorous quote attributed to Fred Jelinek from 1985 reads “Every time I fire a linguist, the performance of our speech recognition system goes up”.</p><p id="b24e" class="pw-post-body-paragraph na nb ho nc b nd ne nf ng nh ni nj nk nl nm nn no np nq nr ns nt nu nv nw nx hh cv"><strong class="nc hp">Speech synthesis</strong> has historically been approached with various stitching mechanisms, but today the state of the art models are large ConvNets (e.g. <a class="bx ps" href="https://deepmind.com/blog/wavenet-launches-google-assistant/" rel="noopener ugc nofollow" target="_blank">WaveNet</a>) that produce raw audio signal outputs.</p><p id="a4e3" class="pw-post-body-paragraph na nb ho nc b nd ne nf ng nh ni nj nk nl nm nn no np nq nr ns nt nu nv nw nx hh cv"><strong class="nc hp">Machine Translation</strong> has usually been approaches with phrase-based statistical techniques, but neural networks are quickly becoming dominant. My favorite architectures are trained in the <a class="bx ps" href="https://arxiv.org/abs/1611.04558" rel="noopener ugc nofollow" target="_blank">multilingual setting</a>, where a single model translates from any source language to any target language, and in weakly supervised (or entirely <a class="bx ps" href="https://arxiv.org/abs/1710.11041" rel="noopener ugc nofollow" target="_blank">unsupervised</a>) settings.</p><p id="386f" class="pw-post-body-paragraph na nb ho nc b nd ne nf ng nh ni nj nk nl nm nn no np nq nr ns nt nu nv nw nx hh cv"><strong class="nc hp">Games. </strong>Explicitly hand-coded Go playing programs have been developed for a long while, but <a class="bx ps" href="https://deepmind.com/blog/alphago-zero-learning-scratch/" rel="noopener ugc nofollow" target="_blank">AlphaGo Zero</a> (a ConvNet that looks at the raw state of the board and plays a move) has now become by far the strongest player of the game. I expect we’re going to see very similar results in other areas, e.g. <a class="bx ps" href="https://blog.openai.com/more-on-dota-2/" rel="noopener ugc nofollow">DOTA 2</a>, or <a class="bx ps" href="https://deepmind.com/blog/deepmind-and-blizzard-open-starcraft-ii-ai-research-environment/" rel="noopener ugc nofollow" target="_blank">StarCraft</a>.</p><p id="089d" class="pw-post-body-paragraph na nb ho nc b nd ne nf ng nh ni nj nk nl nm nn no np nq nr ns nt nu nv nw nx hh cv"><strong class="nc hp">Databases</strong>. More traditional systems outside of Artificial Intelligence are also seeing early hints of a transition. For instance, “<a class="bx ps" href="https://arxiv.org/abs/1712.01208" rel="noopener ugc nofollow" target="_blank">The Case for Learned Index Structures</a>” replaces core components of a data management system with a neural network, outperforming cache-optimized B-Trees by up to 70% in speed while saving an order-of-magnitude in memory.</p><p id="6c91" class="pw-post-body-paragraph na nb ho nc b nd ne nf ng nh ni nj nk nl nm nn no np nq nr ns nt nu nv nw nx hh cv">You’ll notice that many of my links above involve work done at Google. This is because Google is currently at the forefront of re-writing large chunks of itself into Software 2.0 code. “<a class="bx ps" href="https://arxiv.org/abs/1706.05137" rel="noopener ugc nofollow" target="_blank">One model to rule them all</a>” provides an early sketch of what this might look like, where the statistical strength of the individual domains is amalgamated into one consistent understanding of the world.</p><h2 id="0fac" class="op oq ho z or os ot ou ov ow ox oy oz pa pb pc pd pe pf pg ph pi pj pk pl pm cv">The benefits of Software 2.0</h2><p id="d6c4" class="pw-post-body-paragraph na nb ho nc b nd pn nf ng nh po nj nk nl pp nn no np pq nr ns nt pr nv nw nx hh cv">Why should we prefer to port complex programs into Software 2.0? Clearly, one easy answer is that they work better in practice. However, there are a lot of other convenient reasons to prefer this stack. Let’s take a look at some of the benefits of Software 2.0 (think: a ConvNet) compared to Software 1.0 (think: a production-level C++ code base). Software 2.0 is:</p><p id="4486" class="pw-post-body-paragraph na nb ho nc b nd ne nf ng nh ni nj nk nl nm nn no np nq nr ns nt nu nv nw nx hh cv"><strong class="nc hp">Computationally homogeneous</strong>. A typical neural network is, to the first order, made up of a sandwich of only two operations: matrix multiplication and thresholding at zero (ReLU). Compare that with the instruction set of classical software, which is significantly more heterogenous and complex. Because you only have to provide Software 1.0 implementation for a small number of the core computational primitives (e.g. matrix multiply), it is much easier to make various correctness/performance guarantees.</p><p id="061b" class="pw-post-body-paragraph na nb ho nc b nd ne nf ng nh ni nj nk nl nm nn no np nq nr ns nt nu nv nw nx hh cv"><strong class="nc hp">Simple to bake into silicon</strong>. As a corollary, since the instruction set of a neural network is relatively small, it is significantly easier to implement these networks much closer to silicon, e.g. with custom <a class="bx ps" href="https://www.forbes.com/sites/moorinsights/2017/08/04/will-asic-chips-become-the-next-big-thing-in-ai/#7d6d7c0511d9" rel="noopener ugc nofollow" target="_blank">ASICs</a>, <a class="bx ps" href="https://spectrum.ieee.org/semiconductors/design/neuromorphic-chips-are-destined-for-deep-learningor-obscurity" rel="noopener ugc nofollow" target="_blank">neuromorphic chips</a>, and so on. The world will change when low-powered intelligence becomes pervasive around us. E.g., small, inexpensive chips could come with a pretrained ConvNet, a speech recognizer, and a WaveNet speech synthesis network all integrated in a small protobrain that you can attach to stuff.</p><p id="b1d9" class="pw-post-body-paragraph na nb ho nc b nd ne nf ng nh ni nj nk nl nm nn no np nq nr ns nt nu nv nw nx hh cv"><strong class="nc hp">Constant running time</strong>. Every iteration of a typical neural net forward pass takes exactly the same amount of FLOPS. There is zero variability based on the different execution paths your code could take through some sprawling C++ code base. Of course, you could have dynamic compute graphs but the execution flow is normally still significantly constrained. This way we are also almost guaranteed to never find ourselves in unintended infinite loops.</p><p id="c82b" class="pw-post-body-paragraph na nb ho nc b nd ne nf ng nh ni nj nk nl nm nn no np nq nr ns nt nu nv nw nx hh cv"><strong class="nc hp">Constant memory use</strong>. Related to the above, there is no dynamically allocated memory anywhere so there is also little possibility of swapping to disk, or memory leaks that you have to hunt down in your code.</p><p id="3424" class="pw-post-body-paragraph na nb ho nc b nd ne nf ng nh ni nj nk nl nm nn no np nq nr ns nt nu nv nw nx hh cv"><strong class="nc hp">It is highly portable</strong>. A sequence of matrix multiplies is significantly easier to run on arbitrary computational configurations compared to classical binaries or scripts.</p><p id="febb" class="pw-post-body-paragraph na nb ho nc b nd ne nf ng nh ni nj nk nl nm nn no np nq nr ns nt nu nv nw nx hh cv"><strong class="nc hp">It is very agile</strong>. If you had a C++ code and someone wanted you to make it twice as fast (at cost of performance if needed), it would be highly non-trivial to tune the system for the new spec. However, in Software 2.0 we can take our network, remove half of the channels, retrain, and there — it runs exactly at twice the speed and works a bit worse. It’s magic. Conversely, if you happen to get more data/compute, you can immediately make your program work better just by adding more channels and retraining.</p><p id="075d" class="pw-post-body-paragraph na nb ho nc b nd ne nf ng nh ni nj nk nl nm nn no np nq nr ns nt nu nv nw nx hh cv"><strong class="nc hp">Modules can meld into an optimal whole</strong>. Our software is often decomposed into modules that communicate through public functions, APIs, or endpoints. However, if two Software 2.0 modules that were originally trained separately interact, we can easily backpropagate through the whole. Think about how amazing it could be if your web browser could automatically re-design the low-level system instructions 10 stacks down to achieve a higher efficiency in loading web pages. Or if the computer vision library (e.g. OpenCV) you imported could be auto-tuned on your specific data. With 2.0, this is the default behavior.</p><p id="c472" class="pw-post-body-paragraph na nb ho nc b nd ne nf ng nh ni nj nk nl nm nn no np nq nr ns nt nu nv nw nx hh cv"><strong class="nc hp">It is better than you</strong>. Finally, and most importantly, a neural network is a better piece of code than anything you or I can come up with in a large fraction of valuable verticals, which currently at the very least involve anything to do with images/video and sound/speech.</p><h2 id="375e" class="op oq ho z or os ot ou ov ow ox oy oz pa pb pc pd pe pf pg ph pi pj pk pl pm cv">The limitations of Software 2.0</h2><p id="8f48" class="pw-post-body-paragraph na nb ho nc b nd pn nf ng nh po nj nk nl pp nn no np pq nr ns nt pr nv nw nx hh cv">The 2.0 stack also has some of its own disadvantages. At the end of the optimization we’re left with large networks that work well, but it’s very hard to tell how. Across many applications areas, we’ll be left with a choice of using a 90% accurate model we understand, or 99% accurate model we don’t.</p><p id="de95" class="pw-post-body-paragraph na nb ho nc b nd ne nf ng nh ni nj nk nl nm nn no np nq nr ns nt nu nv nw nx hh cv">The 2.0 stack can <a class="bx ps" href="https://motherboard.vice.com/en_us/article/nz7798/weve-already-taught-artificial-intelligence-to-be-racist-sexist" rel="noopener ugc nofollow" target="_blank">fail in unintuitive and embarrassing ways</a> ,or worse, they can “silently fail”, e.g., by silently adopting biases in their training data, which are very difficult to properly analyze and examine when their sizes are easily in the millions in most cases.</p><p id="7188" class="pw-post-body-paragraph na nb ho nc b nd ne nf ng nh ni nj nk nl nm nn no np nq nr ns nt nu nv nw nx hh cv">Finally, we’re still discovering some of the peculiar properties of this stack. For instance, the existence of <a class="bx ps" href="https://blog.openai.com/adversarial-example-research/" rel="noopener ugc nofollow" target="_blank">adversarial examples</a> and <a class="bx ps" href="https://github.com/yenchenlin/awesome-adversarial-machine-learning" rel="noopener ugc nofollow" target="_blank">attacks</a> highlights the unintuitive nature of this stack.</p><h2 id="385e" class="op oq ho z or os ot ou ov ow ox oy oz pa pb pc pd pe pf pg ph pi pj pk pl pm cv">Programming in the 2.0 stack</h2><p id="3494" class="pw-post-body-paragraph na nb ho nc b nd pn nf ng nh po nj nk nl pp nn no np pq nr ns nt pr nv nw nx hh cv">Software 1.0 is code we write. Software 2.0 is code written by the optimization based on an evaluation criterion (such as “classify this training data correctly”). It is likely that any setting where the program is not obvious but one can repeatedly evaluate the performance of it (e.g. — did you classify some images correctly? do you win games of Go?) will be subject to this transition, because the optimization can find much better code than what a human can write.</p><figure class="ob oc od oe of og ny nz paragraph-image"><div class="ny nz pt"><picture><source srcSet="https://miro.medium.com/v2/resize:fit:640/format:webp/1*7aTCueMW8oBRiqkyobunVA.png 640w, https://miro.medium.com/v2/resize:fit:720/format:webp/1*7aTCueMW8oBRiqkyobunVA.png 720w, https://miro.medium.com/v2/resize:fit:750/format:webp/1*7aTCueMW8oBRiqkyobunVA.png 750w, https://miro.medium.com/v2/resize:fit:786/format:webp/1*7aTCueMW8oBRiqkyobunVA.png 786w, https://miro.medium.com/v2/resize:fit:828/format:webp/1*7aTCueMW8oBRiqkyobunVA.png 828w, https://miro.medium.com/v2/resize:fit:1100/format:webp/1*7aTCueMW8oBRiqkyobunVA.png 1100w, https://miro.medium.com/v2/resize:fit:1192/format:webp/1*7aTCueMW8oBRiqkyobunVA.png 1192w" sizes="auto, (min-resolution: 4dppx) and (max-width: 700px) 50vw, (-webkit-min-device-pixel-ratio: 4) and (max-width: 700px) 50vw, (min-resolution: 3dppx) and (max-width: 700px) 67vw, (-webkit-min-device-pixel-ratio: 3) and (max-width: 700px) 65vw, (min-resolution: 2.5dppx) and (max-width: 700px) 80vw, (-webkit-min-device-pixel-ratio: 2.5) and (max-width: 700px) 80vw, (min-resolution: 2dppx) and (max-width: 700px) 100vw, (-webkit-min-device-pixel-ratio: 2) and (max-width: 700px) 100vw, 596px" type="image/webp"/><source data-testid="og" srcSet="https://miro.medium.com/v2/resize:fit:640/1*7aTCueMW8oBRiqkyobunVA.png 640w, https://miro.medium.com/v2/resize:fit:720/1*7aTCueMW8oBRiqkyobunVA.png 720w, https://miro.medium.com/v2/resize:fit:750/1*7aTCueMW8oBRiqkyobunVA.png 750w, https://miro.medium.com/v2/resize:fit:786/1*7aTCueMW8oBRiqkyobunVA.png 786w, https://miro.medium.com/v2/resize:fit:828/1*7aTCueMW8oBRiqkyobunVA.png 828w, https://miro.medium.com/v2/resize:fit:1100/1*7aTCueMW8oBRiqkyobunVA.png 1100w, https://miro.medium.com/v2/resize:fit:1192/1*7aTCueMW8oBRiqkyobunVA.png 1192w" sizes="auto, (min-resolution: 4dppx) and (max-width: 700px) 50vw, (-webkit-min-device-pixel-ratio: 4) and (max-width: 700px) 50vw, (min-resolution: 3dppx) and (max-width: 700px) 67vw, (-webkit-min-device-pixel-ratio: 3) and (max-width: 700px) 65vw, (min-resolution: 2.5dppx) and (max-width: 700px) 80vw, (-webkit-min-device-pixel-ratio: 2.5) and (max-width: 700px) 80vw, (min-resolution: 2dppx) and (max-width: 700px) 100vw, (-webkit-min-device-pixel-ratio: 2) and (max-width: 700px) 100vw, 596px"/><img alt="" class="cs mh ol om" width="596" height="721" loading="lazy" role="presentation"/></picture></div></figure><p id="8133" class="pw-post-body-paragraph na nb ho nc b nd ne nf ng nh ni nj nk nl nm nn no np nq nr ns nt nu nv nw nx hh cv">The lens through which we view trends matters. If you recognize Software 2.0 as a new and emerging programming paradigm instead of simply treating neural networks as a pretty good classifier in the class of machine learning techniques, the extrapolations become more obvious, and it’s clear that there is much more work to do.</p><p id="df86" class="pw-post-body-paragraph na nb ho nc b nd ne nf ng nh ni nj nk nl nm nn no np nq nr ns nt nu nv nw nx hh cv">In particular, we’ve built up a vast amount of tooling that assists humans in writing 1.0 code, such as powerful IDEs with features like syntax highlighting, debuggers, profilers, go to def, git integration, etc. In the 2.0 stack, the programming is done by accumulating, massaging and cleaning datasets. For example, when the network fails in some hard or rare cases, we do not fix those predictions by writing code, but by including more labeled examples of those cases. Who is going to develop the first Software 2.0 IDEs, which help with all of the workflows in accumulating, visualizing, cleaning, labeling, and sourcing datasets? Perhaps the IDE bubbles up images that the network suspects are mislabeled based on the per-example loss, or assists in labeling by seeding labels with predictions, or suggests useful examples to label based on the uncertainty of the network’s predictions.</p><p id="c0b6" class="pw-post-body-paragraph na nb ho nc b nd ne nf ng nh ni nj nk nl nm nn no np nq nr ns nt nu nv nw nx hh cv">Similarly, Github is a very successful home for Software 1.0 code. Is there space for a Software 2.0 Github? In this case repositories are datasets and commits are made up of additions and edits of the labels.</p><p id="5414" class="pw-post-body-paragraph na nb ho nc b nd ne nf ng nh ni nj nk nl nm nn no np nq nr ns nt nu nv nw nx hh cv">Traditional package managers and related serving infrastructure like pip, conda, docker, etc. help us more easily deploy and compose binaries. How do we effectively deploy, share, import and work with Software 2.0 binaries? What is the conda equivalent for neural networks?</p><p id="9cc6" class="pw-post-body-paragraph na nb ho nc b nd ne nf ng nh ni nj nk nl nm nn no np nq nr ns nt nu nv nw nx hh cv">In the short term, Software 2.0 will become increasingly prevalent in any domain where repeated evaluation is possible and cheap, and where the algorithm itself is difficult to design explicitly. There are many exciting opportunities to consider the entire software development ecosystem and how it can be adapted to this new programming paradigm. And in the long run, the future of this paradigm is bright because it is increasingly clear that when we develop AGI, it will certainly be written in Software 2.0.</p></div></div></div></div></section></div></div></article></div><div class="am gt"><div class="gf cs gu gv gw gx"><div class="pu pv am jm do"><div class="am"><a class="bx x by cf ca ab cb ac ae af ag ah ai aj ak" href="https://medium.com/tag/machine-learning?source=post_page---footer_tags--a64152b37c35---------------------------------------" rel="noopener follow"><div class="ei pw dd px py af pz"><span class="z b cr u cv">Machine Learning</span></div></a></div><div class="am"><a class="bx x by cf ca ab cb ac ae af ag ah ai aj ak" href="https://medium.com/tag/artificial-intelligence?source=post_page---footer_tags--a64152b37c35---------------------------------------" rel="noopener follow"><div class="ei pw dd px py af pz"><span class="z b cr u cv">Artificial Intelligence</span></div></a></div><div class="am"><a class="bx x by cf ca ab cb ac ae af ag ah ai aj ak" href="https://medium.com/tag/programming?source=post_page---footer_tags--a64152b37c35---------------------------------------" rel="noopener follow"><div class="ei pw dd px py af pz"><span class="z b cr u cv">Programming</span></div></a></div><div class="am"><a class="bx x by cf ca ab cb ac ae af ag ah ai aj ak" href="https://medium.com/tag/software-development?source=post_page---footer_tags--a64152b37c35---------------------------------------" rel="noopener follow"><div class="ei pw dd px py af pz"><span class="z b cr u cv">Software Development</span></div></a></div><div class="am"><a class="bx x by cf ca ab cb ac ae af ag ah ai aj ak" href="https://medium.com/tag/future?source=post_page---footer_tags--a64152b37c35---------------------------------------" rel="noopener follow"><div class="ei pw dd px py af pz"><span class="z b cr u cv">Future</span></div></a></div></div></div></div><div class="e"></div><footer class="qa qb qc qd qe am j qf qg"><div class="e ce"><div class="am gt"><div class="gf cs gu gv gw gx"><div class="am ip ew"><div class="am j kv"><div class="qh e"><span class="e qi qj qk ql s"><div class="am j kv kw"><div class="pw-multi-vote-icon dm kx ky kz la"><span data-dd-action-name="Susi presentation tracker clap_footer"><a class="bx x by bz ca ab cb ac ae af ag ah ai aj ak" data-testid="footerClapButton" href="https://medium.com/m/signin?actionUrl=https%3A%2F%2Fmedium.com%2F_%2Fvote%2Fp%2Fa64152b37c35&operation=register&redirect=https%3A%2F%2Fkarpathy.medium.com%2Fsoftware-2-0-a64152b37c35&user=Andrej+Karpathy&userId=ac9d9a35533e&source=---footer_actions--a64152b37c35---------------------clap_footer------------------" rel="noopener follow"><div><div class="bt"><div tabindex="-1" class="cq"><div class="lb af lc ld le lf ac lg lh li la" role="presentation"><svg width="24" height="24" viewBox="0 0 24 24" xmlns="http://www.w3.org/2000/svg" aria-label="clap"><path fill-rule="evenodd" clip-rule="evenodd" d="M11.3702 0.827583L12 3.2822L12.6298 0.827583H11.3702Z"></path><path fill-rule="evenodd" clip-rule="evenodd" d="M13.9159 3.95255L15.4386 1.84138L14.2552 1.45158L13.9159 3.95255Z"></path><path fill-rule="evenodd" clip-rule="evenodd" d="M8.58869 1.84055L10.1114 3.95172L9.77379 1.45076L8.58869 1.84055Z"></path><path fill-rule="evenodd" clip-rule="evenodd" d="M18.523 18.9192C17.6623 19.7799 16.7727 20.1664 15.9029 20.25C16.0419 20.1341 16.1776 20.01 16.3101 19.8776C18.6976 17.4891 19.1702 14.9269 17.7087 12.2547L16.7975 10.6516L16.0072 8.9799C15.7465 8.42046 15.8127 7.99921 16.2099 7.69218C16.2866 7.63312 16.3749 7.59082 16.469 7.56801C16.5631 7.5452 16.6609 7.54239 16.7561 7.55977C17.0392 7.60611 17.3023 7.79149 17.4836 8.06046L19.8472 12.2174C20.8229 13.8411 20.9884 16.4538 18.523 18.9192ZM7.52441 18.4822L3.36992 14.3278C3.25591 14.2118 3.1786 14.0648 3.14765 13.9051C3.1167 13.7454 3.13348 13.5802 3.19591 13.43C3.25833 13.2798 3.36363 13.1513 3.49864 13.0606C3.63365 12.9699 3.79239 12.921 3.95503 12.92C4.16606 12.92 4.37875 13.0011 4.54013 13.1617L6.69765 15.3192C6.76738 15.3812 6.85815 15.4142 6.9514 15.4114C7.04465 15.4087 7.13333 15.3704 7.1993 15.3044C7.26526 15.2385 7.30353 15.1498 7.30627 15.0566C7.30902 14.9633 7.27603 14.8725 7.21406 14.8028L5.05655 12.6445L3.60827 11.1962C3.4532 11.0414 3.36595 10.8314 3.36572 10.6124C3.36549 10.3933 3.45229 10.1831 3.60703 10.028C3.76177 9.87297 3.97177 9.78572 4.19084 9.78549C4.40991 9.78526 4.6201 9.87206 4.77517 10.0268L8.21296 13.4662C8.2468 13.5003 8.28703 13.5273 8.33134 13.5457C8.37566 13.5641 8.42317 13.5736 8.47117 13.5736C8.51916 13.5736 8.56668 13.5641 8.61099 13.5457C8.6553 13.5273 8.69553 13.5003 8.72937 13.4662C8.7634 13.4324 8.79041 13.3922 8.80884 13.3479C8.82726 13.3036 8.83675 13.256 8.83675 13.208C8.83675 13.16 8.82726 13.1125 8.80884 13.0682C8.79041 13.0239 8.7634 12.9837 8.72937 12.9498L5.29323 9.51287L4.3233 8.54377C4.16876 8.38892 4.08197 8.17909 4.08197 7.96032C4.08197 7.74155 4.16876 7.53171 4.3233 7.37687C4.4802 7.22625 4.68926 7.14214 4.90675 7.14214C5.12424 7.14214 5.33331 7.22625 5.4902 7.37687L6.46013 8.34514L9.8971 11.7813C9.93077 11.8156 9.97094 11.8429 10.0153 11.8615C10.0596 11.8801 10.1072 11.8897 10.1553 11.8897C10.2034 11.8897 10.251 11.8801 10.2953 11.8615C10.3397 11.8429 10.3798 11.8156 10.4135 11.7813C10.4818 11.7127 10.5202 11.6199 10.5202 11.5231C10.5202 11.4263 10.4818 11.3334 10.4135 11.2649L6.97655 7.82956C6.89975 7.75312 6.83891 7.66217 6.79756 7.56201C6.75622 7.46185 6.7352 7.35447 6.73572 7.24611C6.73507 7.13769 6.75595 7.03021 6.79715 6.92991C6.83835 6.82962 6.89904 6.7385 6.97572 6.66183C7.13084 6.50719 7.34095 6.42035 7.55999 6.42035C7.77903 6.42035 7.98914 6.50719 8.14427 6.66183L13.9308 12.4491C13.9922 12.5076 14.0721 12.5429 14.1567 12.549C14.2413 12.5552 14.3254 12.5318 14.3947 12.4829C14.464 12.434 14.5142 12.3626 14.5367 12.2809C14.5593 12.1991 14.5528 12.112 14.5183 12.0345L13.4011 9.67094C13.1404 9.11149 13.2066 8.69025 13.6047 8.38239C13.6817 8.32396 13.77 8.28204 13.864 8.25926C13.958 8.23648 14.0556 8.23331 14.1509 8.24997C14.4339 8.29632 14.6963 8.4817 14.8783 8.75149L17.0714 12.6114C18.3732 14.9915 17.9545 17.2003 15.7936 19.3611C14.6383 20.5173 13.1917 20.9882 11.6036 20.7283C10.1859 20.4916 8.73765 19.6955 7.52524 18.4822H7.52441ZM10.7503 5.9708L12.8706 8.09108C12.4593 8.59342 12.4047 9.26128 12.7423 9.98128L12.9616 10.4464L9.43861 6.92335C9.37757 6.81002 9.34438 6.68378 9.34179 6.55508C9.34179 6.33411 9.42786 6.12721 9.58344 5.9708C9.74082 5.82112 9.9497 5.73766 10.1669 5.73766C10.3841 5.73766 10.593 5.82112 10.7503 5.9708ZM18.1051 7.67563C17.7948 7.21466 17.3586 6.91756 16.8753 6.83894C16.6812 6.80632 16.4825 6.81388 16.2915 6.86114C16.1005 6.9084 15.9212 6.99436 15.7647 7.1137C15.4535 7.3537 15.2599 7.65659 15.1754 7.99508C14.9266 7.75301 14.611 7.59102 14.2692 7.52997C13.9911 7.4859 13.7061 7.52267 13.4483 7.6359L11.2668 5.45439C10.9747 5.16339 10.5792 5.00001 10.1669 5.00001C9.7546 5.00001 9.35911 5.16339 9.06703 5.45439C8.87539 5.64887 8.73887 5.89078 8.67144 6.15535C8.5271 6.00929 8.35536 5.89315 8.16606 5.81358C7.97675 5.73401 7.77361 5.69257 7.56826 5.69165C7.36292 5.69073 7.15941 5.73034 6.9694 5.8082C6.77938 5.88607 6.60661 6.00066 6.46096 6.14542C6.25382 6.35175 6.10947 6.6126 6.04468 6.8977C5.42068 6.2737 4.39613 6.2737 3.80772 6.86128C3.51632 7.15315 3.35266 7.54872 3.35266 7.96114C3.35266 8.37357 3.51632 8.76914 3.80772 9.06101C3.56937 9.16032 3.30703 9.29935 3.09268 9.5137C2.80164 9.80592 2.63823 10.2015 2.63823 10.614C2.63823 11.0264 2.80164 11.422 3.09268 11.7142L3.6091 12.229C3.33829 12.291 3.08902 12.4243 2.88721 12.6152C2.6854 12.8061 2.53842 13.0476 2.46154 13.3146C2.38467 13.5815 2.38071 13.8642 2.45008 14.1332C2.51945 14.4022 2.65961 14.6477 2.85599 14.8442L7.00965 18.9987C8.32965 20.3178 9.91861 21.1876 11.4852 21.4475C11.806 21.5011 12.1307 21.5282 12.456 21.5286C13.3051 21.5286 14.1087 21.33 14.8494 20.9493C15.0803 20.9824 15.3087 21.0031 15.5346 21.0031C16.8008 21.0031 17.9917 20.4834 19.0403 19.4356C21.8028 16.6731 21.5917 13.7021 20.4786 11.85L18.1051 7.67645V7.67563Z"></path></svg></div></div></div></div></a></span></div><div class="pw-multi-vote-count e lj lk ll lm ln lo lp"><p class="z b y u w"><span class="lq">--</span></p></div></div></span><span class="e ck qm er qn qo"><div class="am j kv kw"><div class="pw-multi-vote-icon dm kx ky kz la"><span data-dd-action-name="Susi presentation tracker clap_footer"><a class="bx x by bz ca ab cb ac ae af ag ah ai aj ak" data-testid="footerClapButton" href="https://medium.com/m/signin?actionUrl=https%3A%2F%2Fmedium.com%2F_%2Fvote%2Fp%2Fa64152b37c35&operation=register&redirect=https%3A%2F%2Fkarpathy.medium.com%2Fsoftware-2-0-a64152b37c35&user=Andrej+Karpathy&userId=ac9d9a35533e&source=---footer_actions--a64152b37c35---------------------clap_footer------------------" rel="noopener follow"><div><div class="bt"><div tabindex="-1" class="cq"><div class="lb af lc ld le lf ac lg lh li la" role="presentation"><svg width="24" height="24" viewBox="0 0 24 24" xmlns="http://www.w3.org/2000/svg" aria-label="clap"><path fill-rule="evenodd" clip-rule="evenodd" d="M11.3702 0.827583L12 3.2822L12.6298 0.827583H11.3702Z"></path><path fill-rule="evenodd" clip-rule="evenodd" d="M13.9159 3.95255L15.4386 1.84138L14.2552 1.45158L13.9159 3.95255Z"></path><path fill-rule="evenodd" clip-rule="evenodd" d="M8.58869 1.84055L10.1114 3.95172L9.77379 1.45076L8.58869 1.84055Z"></path><path fill-rule="evenodd" clip-rule="evenodd" d="M18.523 18.9192C17.6623 19.7799 16.7727 20.1664 15.9029 20.25C16.0419 20.1341 16.1776 20.01 16.3101 19.8776C18.6976 17.4891 19.1702 14.9269 17.7087 12.2547L16.7975 10.6516L16.0072 8.9799C15.7465 8.42046 15.8127 7.99921 16.2099 7.69218C16.2866 7.63312 16.3749 7.59082 16.469 7.56801C16.5631 7.5452 16.6609 7.54239 16.7561 7.55977C17.0392 7.60611 17.3023 7.79149 17.4836 8.06046L19.8472 12.2174C20.8229 13.8411 20.9884 16.4538 18.523 18.9192ZM7.52441 18.4822L3.36992 14.3278C3.25591 14.2118 3.1786 14.0648 3.14765 13.9051C3.1167 13.7454 3.13348 13.5802 3.19591 13.43C3.25833 13.2798 3.36363 13.1513 3.49864 13.0606C3.63365 12.9699 3.79239 12.921 3.95503 12.92C4.16606 12.92 4.37875 13.0011 4.54013 13.1617L6.69765 15.3192C6.76738 15.3812 6.85815 15.4142 6.9514 15.4114C7.04465 15.4087 7.13333 15.3704 7.1993 15.3044C7.26526 15.2385 7.30353 15.1498 7.30627 15.0566C7.30902 14.9633 7.27603 14.8725 7.21406 14.8028L5.05655 12.6445L3.60827 11.1962C3.4532 11.0414 3.36595 10.8314 3.36572 10.6124C3.36549 10.3933 3.45229 10.1831 3.60703 10.028C3.76177 9.87297 3.97177 9.78572 4.19084 9.78549C4.40991 9.78526 4.6201 9.87206 4.77517 10.0268L8.21296 13.4662C8.2468 13.5003 8.28703 13.5273 8.33134 13.5457C8.37566 13.5641 8.42317 13.5736 8.47117 13.5736C8.51916 13.5736 8.56668 13.5641 8.61099 13.5457C8.6553 13.5273 8.69553 13.5003 8.72937 13.4662C8.7634 13.4324 8.79041 13.3922 8.80884 13.3479C8.82726 13.3036 8.83675 13.256 8.83675 13.208C8.83675 13.16 8.82726 13.1125 8.80884 13.0682C8.79041 13.0239 8.7634 12.9837 8.72937 12.9498L5.29323 9.51287L4.3233 8.54377C4.16876 8.38892 4.08197 8.17909 4.08197 7.96032C4.08197 7.74155 4.16876 7.53171 4.3233 7.37687C4.4802 7.22625 4.68926 7.14214 4.90675 7.14214C5.12424 7.14214 5.33331 7.22625 5.4902 7.37687L6.46013 8.34514L9.8971 11.7813C9.93077 11.8156 9.97094 11.8429 10.0153 11.8615C10.0596 11.8801 10.1072 11.8897 10.1553 11.8897C10.2034 11.8897 10.251 11.8801 10.2953 11.8615C10.3397 11.8429 10.3798 11.8156 10.4135 11.7813C10.4818 11.7127 10.5202 11.6199 10.5202 11.5231C10.5202 11.4263 10.4818 11.3334 10.4135 11.2649L6.97655 7.82956C6.89975 7.75312 6.83891 7.66217 6.79756 7.56201C6.75622 7.46185 6.7352 7.35447 6.73572 7.24611C6.73507 7.13769 6.75595 7.03021 6.79715 6.92991C6.83835 6.82962 6.89904 6.7385 6.97572 6.66183C7.13084 6.50719 7.34095 6.42035 7.55999 6.42035C7.77903 6.42035 7.98914 6.50719 8.14427 6.66183L13.9308 12.4491C13.9922 12.5076 14.0721 12.5429 14.1567 12.549C14.2413 12.5552 14.3254 12.5318 14.3947 12.4829C14.464 12.434 14.5142 12.3626 14.5367 12.2809C14.5593 12.1991 14.5528 12.112 14.5183 12.0345L13.4011 9.67094C13.1404 9.11149 13.2066 8.69025 13.6047 8.38239C13.6817 8.32396 13.77 8.28204 13.864 8.25926C13.958 8.23648 14.0556 8.23331 14.1509 8.24997C14.4339 8.29632 14.6963 8.4817 14.8783 8.75149L17.0714 12.6114C18.3732 14.9915 17.9545 17.2003 15.7936 19.3611C14.6383 20.5173 13.1917 20.9882 11.6036 20.7283C10.1859 20.4916 8.73765 19.6955 7.52524 18.4822H7.52441ZM10.7503 5.9708L12.8706 8.09108C12.4593 8.59342 12.4047 9.26128 12.7423 9.98128L12.9616 10.4464L9.43861 6.92335C9.37757 6.81002 9.34438 6.68378 9.34179 6.55508C9.34179 6.33411 9.42786 6.12721 9.58344 5.9708C9.74082 5.82112 9.9497 5.73766 10.1669 5.73766C10.3841 5.73766 10.593 5.82112 10.7503 5.9708ZM18.1051 7.67563C17.7948 7.21466 17.3586 6.91756 16.8753 6.83894C16.6812 6.80632 16.4825 6.81388 16.2915 6.86114C16.1005 6.9084 15.9212 6.99436 15.7647 7.1137C15.4535 7.3537 15.2599 7.65659 15.1754 7.99508C14.9266 7.75301 14.611 7.59102 14.2692 7.52997C13.9911 7.4859 13.7061 7.52267 13.4483 7.6359L11.2668 5.45439C10.9747 5.16339 10.5792 5.00001 10.1669 5.00001C9.7546 5.00001 9.35911 5.16339 9.06703 5.45439C8.87539 5.64887 8.73887 5.89078 8.67144 6.15535C8.5271 6.00929 8.35536 5.89315 8.16606 5.81358C7.97675 5.73401 7.77361 5.69257 7.56826 5.69165C7.36292 5.69073 7.15941 5.73034 6.9694 5.8082C6.77938 5.88607 6.60661 6.00066 6.46096 6.14542C6.25382 6.35175 6.10947 6.6126 6.04468 6.8977C5.42068 6.2737 4.39613 6.2737 3.80772 6.86128C3.51632 7.15315 3.35266 7.54872 3.35266 7.96114C3.35266 8.37357 3.51632 8.76914 3.80772 9.06101C3.56937 9.16032 3.30703 9.29935 3.09268 9.5137C2.80164 9.80592 2.63823 10.2015 2.63823 10.614C2.63823 11.0264 2.80164 11.422 3.09268 11.7142L3.6091 12.229C3.33829 12.291 3.08902 12.4243 2.88721 12.6152C2.6854 12.8061 2.53842 13.0476 2.46154 13.3146C2.38467 13.5815 2.38071 13.8642 2.45008 14.1332C2.51945 14.4022 2.65961 14.6477 2.85599 14.8442L7.00965 18.9987C8.32965 20.3178 9.91861 21.1876 11.4852 21.4475C11.806 21.5011 12.1307 21.5282 12.456 21.5286C13.3051 21.5286 14.1087 21.33 14.8494 20.9493C15.0803 20.9824 15.3087 21.0031 15.5346 21.0031C16.8008 21.0031 17.9917 20.4834 19.0403 19.4356C21.8028 16.6731 21.5917 13.7021 20.4786 11.85L18.1051 7.67645V7.67563Z"></path></svg></div></div></div></div></a></span></div><div class="pw-multi-vote-count e lj lk ll lm ln lo lp"><p class="z b y u w"><span class="lq">--</span></p></div></div></span></div><div class="qp am"><div><div class="bt"><div tabindex="-1" class="cq"><button class="af lb fe lu am j dn lv lw" aria-label="responses"><svg width="24" height="24" viewBox="0 0 24 24" xmlns="http://www.w3.org/2000/svg" class="lt"><path d="M18.0064 16.8029C19.5393 15.347 20.24 13.4782 20.24 11.4818C20.24 7.35659 16.7088 4 12.1914 4C7.67395 4 4 7.35659 4 11.4818C4 15.6077 7.67395 18.9643 12.1914 18.9643C13.0081 18.9643 13.8134 18.8527 14.5839 18.6374C14.815 18.8368 15.064 19.0281 15.3284 19.1956C16.3876 19.8892 17.5312 20.24 18.7266 20.24C18.9505 20.232 19.1273 20.1284 19.2132 19.953C19.2554 19.8711 19.2738 19.7795 19.2664 19.688C19.2589 19.5966 19.2259 19.5089 19.1711 19.4347C18.6763 18.765 18.326 18.0714 18.1313 17.3777C18.0754 17.1816 18.0337 16.9817 18.0064 16.7798V16.8029ZM14.884 17.8585L14.8167 17.6353L14.502 17.7309C13.7529 17.9559 12.9738 18.0698 12.1906 18.069C8.16787 18.069 4.89943 15.1142 4.89943 11.4818C4.89943 7.84932 8.16787 4.89376 12.1906 4.89376C16.2051 4.89376 19.3033 7.85171 19.3033 11.4874C19.3033 13.2813 18.695 14.9556 17.2757 16.2073L17.081 16.3747V16.6299C17.081 16.6857 17.0811 16.7814 17.0973 16.9249C17.1216 17.1561 17.1784 17.4032 17.2514 17.6584C17.4055 18.2165 17.6488 18.7746 17.9732 19.3167C17.2189 19.2051 16.489 18.9261 15.8077 18.4716C15.532 18.2962 15.094 18.0889 14.8669 17.8816L14.884 17.8585Z"></path></svg><p class="z b cr u w"><span class="pw-responses-count ls lt">192</span></p></button></div></div></div></div><div class="am j w"><div><div class="bt"><div tabindex="-1" class="cq"><span data-dd-action-name="Susi presentation tracker repost_footer"><a class="bx x by bz ca ab cb ac ae af ag ah ai aj ak" href="https://medium.com/m/signin?actionUrl=https%3A%2F%2Fmedium.com%2F_%2Frepost%2Fp%2Fa64152b37c35&operation=register&redirect=https%3A%2F%2Fkarpathy.medium.com%2Fsoftware-2-0-a64152b37c35&user=Andrej+Karpathy&userId=ac9d9a35533e&source=---footer_actions--a64152b37c35---------------------repost_footer------------------" rel="noopener follow"><span class="lb dz bx af fe lg am j lh li la ac lx ly" aria-disabled="false" aria-label="Repost"><div class="dg df e dm"><svg xmlns="http://www.w3.org/2000/svg" width="24" height="24" viewBox="0 0 24 24" fill="none"><path d="M3 8.72727L5.64706 6M5.64706 6L8.29412 8.72727M5.64706 6V14.1818V15.2475C5.64706 15.9633 5.92589 16.6509 6.42442 17.1646C6.94277 17.6986 7.6553 18 8.39955 18H12" stroke="currentColor" stroke-linecap="round"></path><path d="M21 15.2727L18.3529 18M18.3529 18L15.7059 15.2727M18.3529 18V9.81818V8.75249C18.3529 8.03671 18.0741 7.34908 17.5756 6.83545C17.0572 6.30139 16.3447 6 15.6004 6L12 6" stroke="currentColor" stroke-linecap="round"></path></svg></div></span></a></span></div></div></div><div class="ls e"></div></div></div><div class="am j"><div class="ck"><div class="qq e ej"><div><div class="bt"><div tabindex="-1" class="cq"><span data-dd-action-name="Susi presentation tracker bookmark_footer"><a class="bx x by bz ca ab cb ac ae af ag ah ai aj ak" data-testid="footerBookmarkButton" href="https://medium.com/m/signin?actionUrl=https%3A%2F%2Fmedium.com%2F_%2Fbookmark%2Fp%2Fa64152b37c35&operation=register&redirect=https%3A%2F%2Fkarpathy.medium.com%2Fsoftware-2-0-a64152b37c35&source=---footer_actions--a64152b37c35---------------------bookmark_footer------------------" rel="noopener follow"><span class=""><svg width="25" height="25" viewBox="0 0 25 25" fill="none" xmlns="http://www.w3.org/2000/svg" class="w lz" aria-label="Add to list bookmark button"><path d="M18 2.5C18 2.22386 18.2239 2 18.5 2C18.7762 2 19 2.22386 19 2.5V5H21.5C21.7761 5 22 5.22386 22 5.5C22 5.77614 21.7761 6 21.5 6H19V8.5C19 8.77614 18.7762 9 18.5 9C18.2239 9 18 8.77614 18 8.5V6H15.5C15.2239 6 15 5.77614 15 5.5C15 5.22386 15.2239 5 15.5 5H18V2.5Z" fill="currentColor"></path><path d="M7 7C7 6.44772 7.44772 6 8 6H11.5C11.7761 6 12 5.77614 12 5.5C12 5.22386 11.7761 5 11.5 5H8C6.89543 5 6 5.89543 6 7V21C6 21.1905 6.10823 21.3644 6.27912 21.4486C6.45002 21.5327 6.65387 21.5125 6.80486 21.3963L12.5 17L18.1951 21.3963C18.3461 21.5125 18.55 21.5327 18.7209 21.4486C18.8918 21.3644 19 21.1905 19 21V12.5C19 12.2239 18.7761 12 18.5 12C18.2239 12 18 12.2239 18 12.5V19.9846L12.8049 15.9729C12.6251 15.8346 12.3749 15.8346 12.1951 15.9729L7 19.9846V7Z" fill="currentColor"></path></svg></span></a></span></div></div></div></div></div><div class="qq e ej"><div class="bt" aria-describedby="postFooterSocialMenu" aria-labelledby="postFooterSocialMenu"><div><div class="bt"><div tabindex="-1" class="cq"><button aria-controls="postFooterSocialMenu" aria-expanded="false" aria-label="Share Post" data-testid="footerSocialShareButton" class="bx dn by bz ca ab cb mi ae af ag bm mj mk lw ml"><svg width="24" height="24" viewBox="0 0 24 24" fill="none" xmlns="http://www.w3.org/2000/svg"><path fill-rule="evenodd" clip-rule="evenodd" d="M15.2176 4.93118C15.1883 4.98289 15.1484 5.0278 15.1004 5.06293L15.1118 5.06859C15.024 5.12481 14.9269 5.15261 14.8204 5.14293C14.7004 5.13293 14.6004 5.08293 14.5204 5.01293L12.5004 2.99293V10.0629C12.5004 10.3429 12.2704 10.5629 12.0004 10.5629C11.7304 10.5629 11.5004 10.3429 11.5004 10.0629V3.02293L9.50042 5.02293C9.42474 5.09456 9.32654 5.13771 9.22259 5.14501C9.11863 5.1523 9.01537 5.12329 8.93042 5.06293H8.91042C8.82042 4.99293 8.75042 4.88293 8.75042 4.76293C8.74294 4.70553 8.74801 4.64719 8.76528 4.59195C8.78254 4.5367 8.81159 4.48586 8.85042 4.44293L11.6504 1.64293C11.7439 1.55132 11.8695 1.5 12.0004 1.5C12.1313 1.5 12.257 1.55132 12.3504 1.64293L15.1504 4.43293C15.1931 4.47602 15.2258 4.52782 15.2466 4.58479C15.2673 4.64175 15.2754 4.70252 15.2704 4.76293C15.2649 4.82212 15.2469 4.87946 15.2176 4.93118ZM15.1118 5.06859L15.1204 5.07293V5.06293C15.1175 5.06485 15.1147 5.06674 15.1118 5.06859ZM16 7.06293H17.5C18.0304 7.06293 18.5391 7.27364 18.9142 7.64871C19.2893 8.02379 19.5 8.53249 19.5 9.06293V19.0629C19.5 19.5934 19.2893 20.1021 18.9142 20.4771C18.5391 20.8522 18.0304 21.0629 17.5 21.0629H6.5C5.4 21.0629 4.5 20.1629 4.5 19.0629V9.06293C4.5 8.53249 4.71071 8.02379 5.08579 7.64871C5.46086 7.27364 5.96957 7.06293 6.5 7.06293H8C8.13188 7.06481 8.25769 7.11872 8.35 7.21293C8.44421 7.30524 8.49812 7.43104 8.5 7.56293C8.49812 7.69481 8.44421 7.82061 8.35 7.91293C8.25769 8.00713 8.13188 8.06105 8 8.06293H6.4C5.9 8.06293 5.5 8.46293 5.5 8.96293V19.1629C5.5 19.4016 5.59482 19.6305 5.7636 19.7993C5.93239 19.9681 6.16131 20.0629 6.4 20.0629H17.6C18.1 20.0629 18.5 19.6629 18.5 19.1629V8.96293C18.5 8.46293 18.1 8.06293 17.6 8.06293H16C15.8674 8.06293 15.7402 8.01025 15.6464 7.91648C15.5527 7.82271 15.5 7.69554 15.5 7.56293C15.5 7.43032 15.5527 7.30314 15.6464 7.20937C15.7402 7.11561 15.8674 7.06293 16 7.06293Z" fill="currentColor"></path></svg></button></div></div></div></div></div></div></div></div></div></div></footer><div class="qr e"><div class="am gt"><div class="gf cs gu gv gw gx"><div class="am iz ix iv qs qt"><div class="qu qv qw qx qy qz ra rb rc rd am ip"><div class="ck cz"><a tabindex="0" rel="noopener follow" href="/?source=post_page---post_author_info--a64152b37c35---------------------------------------" data-discover="true"><div class="e dm"><img alt="Andrej Karpathy" class="e bs ee re rf ei" src="https://miro.medium.com/v2/resize:fill:96:96/0*8ldFdx9B6FhSkQmV.jpeg" width="48" height="48" loading="lazy"/><div class="ed ee e re rf dr g cf rg"></div></div></a></div><div class="da r s"><a tabindex="0" rel="noopener follow" href="/?source=post_page---post_author_info--a64152b37c35---------------------------------------" data-discover="true"><div class="e dm"><img alt="Andrej Karpathy" class="e bs ee rh ri ei" src="https://miro.medium.com/v2/resize:fill:128:128/0*8ldFdx9B6FhSkQmV.jpeg" width="64" height="64" loading="lazy"/><div class="ed ee e rh ri dr g cf rg"></div></div></a></div><div class="da r s rj ej"><div class="am"></div></div></div><div class="am ex el"><div class="rk rl rm rn ro e"><a class="bx x by ca ab cb ac ae af ag ah ai aj ak am j" rel="noopener follow" href="/?source=post_page---post_author_info--a64152b37c35---------------------------------------" data-discover="true"><h2 class="pw-author-name z rq rr rs rt ru rv rw nl rx ry np rz sa nt sb sc cv"><span class="hh rp">Written by <!-- -->Andrej Karpathy</span></h2></a><div class="sd am jb"><div class="e ej"><span class="pw-follower-count z b cr u w"><a class="bx x by bz ca ab cb ac ae af ag ah ai jk" rel="noopener follow" href="/followers?source=post_page---post_author_info--a64152b37c35---------------------------------------" data-discover="true">61K followers</a></span></div><div class="z b cr u w am se"><span class="sf e" aria-hidden="true"><span class="z b cr u w">·</span></span><a class="bx x by bz ca ab cb ac ae af ag ah ai jk" rel="noopener follow" href="/following?source=post_page---post_author_info--a64152b37c35---------------------------------------" data-discover="true">186 following</a></div></div><div class="sg e"><p class="z b cr u cv"><span class="hh">I like to train deep neural nets on large datasets.</span></p></div></div></div><div class="ck cz"><div class="am"></div></div></div></div></div></div><div><div class="am gt"><div class="gf cs gu gv gw gx"><div class="sh si"><div class="am kv"><div class="sj ef eg e ei"></div><div class="v sk e"><div class="sl sm sn e ei"></div><div class="sm so e ei"></div></div></div><div class="sp e"><div class="sl sq cs e ei"></div><div class="sl sq cs e ei"></div><div class="sq sr e ei"></div></div></div></div></div></div><div><div class="ss st su sv sw e"><div class="sx cs k sy sz gq gr gs"></div><div class="am gt"><div class="gf cs gu gv gw gx"><div class="sl ta tb e ei"></div><div class="ta tc e ei"></div><div class="td am te"><div class="tf cm e ei"></div><div class="tf cm e ei"></div></div><div class="tg ta rf e ei"></div></div></div></div></div><div class="th ti tj tk tl e"><div><div class="am gt"><div class="gf cs gu gv gw gx"><div class="tm am kv jm tn to tp tq tr ts tt tu tv tw tx ty tz ua ub"><div class="uc ud ue ro uf ug uh rn ui uj uk ul um un uo up uq ur us ut uu"><div class="tg uv cs e ei"></div><div class="uw ef ux e ei"></div><div class="sl uy uz e ei"></div><div class="sl uy va e ei"></div><div class="vb uy vc e ei"></div></div><div class="uc ud ue ro uf ug uh rn ui uj uk ul um un uo up uq ur us ut uu"><div class="tg uv cs e ei"></div><div class="uw ef ux e ei"></div><div class="sl uy uz e ei"></div><div class="sl uy va e ei"></div><div class="vb uy vc e ei"></div></div></div></div></div></div><div><div class="am gt"><div class="gf cs gu gv gw gx"><div class="tm am kv jm tn to tp tq tr ts tt tu tv tw tx ty tz ua ub"><div class="uc ud ue ro uf ug uh rn ui uj uk ul um un uo up uq ur us ut uu"><div class="tg uv cs e ei"></div><div class="uw ef ux e ei"></div><div class="sl uy uz e ei"></div><div class="sl uy va e ei"></div><div class="vb uy vc e ei"></div></div><div class="uc ud ue ro uf ug uh rn ui uj uk ul um un uo up uq ur us ut uu"><div class="tg uv cs e ei"></div><div class="uw ef ux e ei"></div><div class="sl uy uz e ei"></div><div class="sl uy va e ei"></div><div class="vb uy vc e ei"></div></div></div></div></div></div><div class="ck cz da"><div class="sx cs vd ve"></div><div class="am gt"><div class="gf cs gu gv gw gx"><div class="vf am kv jm"><div class="vg vh e"><a class="bx x by bz ca ab cb ac ae af ag ah ai aj ak" href="https://help.medium.com/hc/en-us?source=post_page-----a64152b37c35---------------------------------------" rel="noopener follow"><p class="z b y u w">Help</p></a></div><div class="vg vh e"><a class="bx x by bz ca ab cb ac ae af ag ah ai aj ak" href="https://status.medium.com/?source=post_page-----a64152b37c35---------------------------------------" rel="noopener follow"><p class="z b y u w">Status</p></a></div><div class="vg vh e"><a class="bx x by bz ca ab cb ac ae af ag ah ai aj ak" href="https://medium.com/about?autoplay=1&source=post_page-----a64152b37c35---------------------------------------" rel="noopener follow"><p class="z b y u w">About</p></a></div><div class="vg vh e"><a class="bx x by bz ca ab cb ac ae af ag ah ai aj ak" href="https://medium.com/jobs-at-medium/work-at-medium-959d1a85284e?source=post_page-----a64152b37c35---------------------------------------" rel="noopener follow"><p class="z b y u w">Careers</p></a></div><div class="vg vh e"><a class="bx x by bz ca ab cb ac ae af ag ah ai aj ak" href="mailto:pressinquiries@medium.com" rel="noopener follow"><p class="z b y u w">Press</p></a></div><div class="vg vh e"><a class="bx x by bz ca ab cb ac ae af ag ah ai aj ak" href="https://blog.medium.com/?source=post_page-----a64152b37c35---------------------------------------" rel="noopener follow"><p class="z b y u w">Blog</p></a></div><div class="vg vh e"><a class="bx x by bz ca ab cb ac ae af ag ah ai aj ak" href="https://medium.com/store" rel="noopener follow" target="_blank"><p class="z b y u w">Store</p></a></div><div class="vg vh e"><a class="bx x by bz ca ab cb ac ae af ag ah ai aj ak" href="https://policy.medium.com/medium-privacy-policy-f03bf92035c9?source=post_page-----a64152b37c35---------------------------------------" rel="noopener follow"><p class="z b y u w">Privacy</p></a></div><div class="vg vh e"><a class="bx x by bz ca ab cb ac ae af ag ah ai aj ak" href="https://policy.medium.com/medium-rules-30e5502c4eb4?source=post_page-----a64152b37c35---------------------------------------" rel="noopener follow"><p class="z b y u w">Rules</p></a></div><div class="vg vh e"><a class="bx x by bz ca ab cb ac ae af ag ah ai aj ak" href="https://policy.medium.com/medium-terms-of-service-9db0094a1e0f?source=post_page-----a64152b37c35---------------------------------------" rel="noopener follow"><p class="z b y u w">Terms</p></a></div><div class="vg e"><a class="bx x by bz ca ab cb ac ae af ag ah ai aj ak" href="https://speechify.com/medium?source=post_page-----a64152b37c35---------------------------------------" rel="noopener follow"><p class="z b y u w">Text to speech</p></a></div></div></div></div></div></div></div></div></div></div><script>((storageKey2, restoreKey) => {
if (!window.history.state || !window.history.state.key) {
let key = Math.random().toString(32).slice(2);
window.history.replaceState({ key }, "");
}
try {
let positions = JSON.parse(sessionStorage.getItem(storageKey2) || "{}");
let storedY = positions[restoreKey || window.history.state.key];
if (typeof storedY === "number") {
window.scrollTo(0, storedY);
}
} catch (error) {
console.error(error);
sessionStorage.removeItem(storageKey2);
}
})("react-router-scroll-positions", null)</script><script>window.__reactRouterContext = {"basename":"/","future":{"unstable_optimizeDeps":true,"v8_passThroughRequests":false,"unstable_trailingSlashAwareDataRequests":false,"unstable_previewServerPrerendering":false,"v8_middleware":true,"v8_splitRouteModules":false,"v8_viteEnvironmentApi":false},"routeDiscovery":{"mode":"lazy","manifestPath":"/__manifest"},"ssr":true,"isSpaMode":false};window.__reactRouterContext.stream = new ReadableStream({start(controller){window.__reactRouterContext.streamController = controller;}}).pipeThrough(new TextEncoderStream());</script><script type="module" async="">;
import * as route0 from "https://cdn-client.medium.com/lite/assets/root-BMBCWO_EnW.js";
import * as route1 from "https://cdn-client.medium.com/lite/assets/ShowEntitySlug-DrZts1oV_Z.js";
window.__reactRouterManifest = {
"entry": {
"module": "https://cdn-client.medium.com/lite/assets/entry.client-DtJLrZR8VY.js",
"imports": [
"https://cdn-client.medium.com/lite/assets/vendor-react-CpuriD7yOR.js",
"https://cdn-client.medium.com/lite/assets/RoutingContext-N-dlYRSyd6.js",
"https://cdn-client.medium.com/lite/assets/index.esm-dkQXnoW74C.js",
"https://cdn-client.medium.com/lite/assets/vendor-apollo-B0FLDVoLJa.js",
"https://cdn-client.medium.com/lite/assets/catalog-B5naEjnKWy.js",
"https://cdn-client.medium.com/lite/assets/vendor-lodash-C0lcM7XZQI.js",
"https://cdn-client.medium.com/lite/assets/useClientViewerOrVisitorId-CBF3XYp5m3.js",
"https://cdn-client.medium.com/lite/assets/Logo-CAikMcA5xF.js",
"https://cdn-client.medium.com/lite/assets/entityPageRegistry-BzOK2jlsWT.js",
"https://cdn-client.medium.com/lite/assets/actions-BuaTAfK8Ws.js",
"https://cdn-client.medium.com/lite/assets/performance-B_nPciEuBN.js",
"https://cdn-client.medium.com/lite/assets/useragent-DM_y0Y6BQs.js",
"https://cdn-client.medium.com/lite/assets/actions-DarRc50fcc.js",
"https://cdn-client.medium.com/lite/assets/actions-DGO5CiB3v-.js",
"https://cdn-client.medium.com/lite/assets/actions-BBSCsZ7iAN.js",
"https://cdn-client.medium.com/lite/assets/alternateUserTopBannerActions-CFUxDhHrPM.js",
"https://cdn-client.medium.com/lite/assets/googleOneTapActions-D-Kvzy3lkp.js",
"https://cdn-client.medium.com/lite/assets/actions-1GHcCLwm6f.js",
"https://cdn-client.medium.com/lite/assets/action-CLBCv6uGgP.js",
"https://cdn-client.medium.com/lite/assets/vendor-fela-Cy19ySZi4y.js",
"https://cdn-client.medium.com/lite/assets/color-DcO1XOeLw1.js"
],
"css": []
},
"routes": {
"root": {
"id": "root",
"path": "",
"hasAction": false,
"hasLoader": false,
"hasClientAction": false,
"hasClientLoader": false,
"hasClientMiddleware": false,
"hasDefaultExport": true,
"hasErrorBoundary": true,
"module": "https://cdn-client.medium.com/lite/assets/root-BMBCWO_EnW.js",
"imports": [
"https://cdn-client.medium.com/lite/assets/vendor-react-CpuriD7yOR.js",
"https://cdn-client.medium.com/lite/assets/RoutingContext-N-dlYRSyd6.js",
"https://cdn-client.medium.com/lite/assets/index.esm-dkQXnoW74C.js",
"https://cdn-client.medium.com/lite/assets/vendor-apollo-B0FLDVoLJa.js",
"https://cdn-client.medium.com/lite/assets/catalog-B5naEjnKWy.js",
"https://cdn-client.medium.com/lite/assets/vendor-lodash-C0lcM7XZQI.js",
"https://cdn-client.medium.com/lite/assets/useClientViewerOrVisitorId-CBF3XYp5m3.js",
"https://cdn-client.medium.com/lite/assets/Logo-CAikMcA5xF.js",
"https://cdn-client.medium.com/lite/assets/entityPageRegistry-BzOK2jlsWT.js",
"https://cdn-client.medium.com/lite/assets/actions-BuaTAfK8Ws.js",
"https://cdn-client.medium.com/lite/assets/performance-B_nPciEuBN.js",
"https://cdn-client.medium.com/lite/assets/useragent-DM_y0Y6BQs.js",
"https://cdn-client.medium.com/lite/assets/actions-DarRc50fcc.js",
"https://cdn-client.medium.com/lite/assets/actions-DGO5CiB3v-.js",
"https://cdn-client.medium.com/lite/assets/actions-BBSCsZ7iAN.js",
"https://cdn-client.medium.com/lite/assets/alternateUserTopBannerActions-CFUxDhHrPM.js",
"https://cdn-client.medium.com/lite/assets/googleOneTapActions-D-Kvzy3lkp.js",
"https://cdn-client.medium.com/lite/assets/actions-1GHcCLwm6f.js",
"https://cdn-client.medium.com/lite/assets/action-CLBCv6uGgP.js",
"https://cdn-client.medium.com/lite/assets/vendor-fela-Cy19ySZi4y.js",
"https://cdn-client.medium.com/lite/assets/color-DcO1XOeLw1.js",
"https://cdn-client.medium.com/lite/assets/Redirect-CNnLIdBBDa.js",
"https://cdn-client.medium.com/lite/assets/useBaseCacheControl.generated.graphql-mg_C73LFNG.js",
"https://cdn-client.medium.com/lite/assets/selectors-Civm4lBZJ3.js",
"https://cdn-client.medium.com/lite/assets/NotFoundHandler-D2FdvSX1Qb.js",
"https://cdn-client.medium.com/lite/assets/GoogleAnalytics-CpX84ZJVzm.js",
"https://cdn-client.medium.com/lite/assets/WithClientFlag-D6H5iZ5feY.js",
"https://cdn-client.medium.com/lite/assets/Layer-BoAybuFPli.js",
"https://cdn-client.medium.com/lite/assets/visits-BnXWJEFetB.js",
"https://cdn-client.medium.com/lite/assets/useClientResponsiveSize-DMj2FEayo4.js",
"https://cdn-client.medium.com/lite/assets/UserIdStorage-CucM5frZ7R.js",
"https://cdn-client.medium.com/lite/assets/ErrorScreen-DfI8g0yVkR.js",
"https://cdn-client.medium.com/lite/assets/index-DHOhXfp0T0.js",
"https://cdn-client.medium.com/lite/assets/SusiClickable-D3bZIDOhCt.js",
"https://cdn-client.medium.com/lite/assets/SusiModalContainer-DTz4QdX9F_.js",
"https://cdn-client.medium.com/lite/assets/CloseButton-C6-EVtDglw.js",
"https://cdn-client.medium.com/lite/assets/useReduxSusi-BDnRTMQqz7.js",
"https://cdn-client.medium.com/lite/assets/RememberMeCheckbox-C1ARm7l_lx.js",
"https://cdn-client.medium.com/lite/assets/GoogleRecaptcha-kaX0womyEh.js",
"https://cdn-client.medium.com/lite/assets/FacebookSusiButton-Cdj6Ecnpls.js",
"https://cdn-client.medium.com/lite/assets/TextField-BswpTNUiLv.js",
"https://cdn-client.medium.com/lite/assets/email-x6DqV6Od7s.js",
"https://cdn-client.medium.com/lite/assets/CircularImage-ChC4HYcXtN.js",
"https://cdn-client.medium.com/lite/assets/ZoomableImage-CMLNdmn9hp.js",
"https://cdn-client.medium.com/lite/assets/LoadingSpinner-C-Fc1l7DXa.js",
"https://cdn-client.medium.com/lite/assets/Dialog-AkVTR0-Mir.js",
"https://cdn-client.medium.com/lite/assets/SusiModal.generated.graphql-C2CdXisXOb.js",
"https://cdn-client.medium.com/lite/assets/Toast-BeYgQ7UP8I.js",
"https://cdn-client.medium.com/lite/assets/VisibilityController-DfDM1mH55T.js",
"https://cdn-client.medium.com/lite/assets/Popover-BEHTzJA9dJ.js",
"https://cdn-client.medium.com/lite/assets/miroImage-DnVurRI6Te.js",
"https://cdn-client.medium.com/lite/assets/glyph-DCVZQIrAg0.js",
"https://cdn-client.medium.com/lite/assets/isPost-Jkp00EWxnS.js",
"https://cdn-client.medium.com/lite/assets/usePreference-D22I6RuVgo.js",
"https://cdn-client.medium.com/lite/assets/useExperimentExposureLogger-RvbVwpzFqo.js",
"https://cdn-client.medium.com/lite/assets/useGateExposureLogger-CNFBfDLsZy.js",
"https://cdn-client.medium.com/lite/assets/NotFoundScreen-Cbxs6enf11.js",
"https://cdn-client.medium.com/lite/assets/LoadingScreen-DXn01VrTWl.js",
"https://cdn-client.medium.com/lite/assets/index-9bMYTUUBe-.js",
"https://cdn-client.medium.com/lite/assets/GlobalNavigationContext-CFsD4sSJxy.js",
"https://cdn-client.medium.com/lite/assets/useIsomorphicLayoutEffect-mTFvRtpWc0.js",
"https://cdn-client.medium.com/lite/assets/OpenInApp-DfYOBi2m2t.js",
"https://cdn-client.medium.com/lite/assets/useGoogleTagEvent-BVvaY-rQcL.js",
"https://cdn-client.medium.com/lite/assets/SettingsPage-1nQFOpO_S1.js",
"https://cdn-client.medium.com/lite/assets/PageHeader-vZiraAAZX8.js",
"https://cdn-client.medium.com/lite/assets/PageNavigationBar-BE30YyYyoY.js",
"https://cdn-client.medium.com/lite/assets/ResizeObserver-R46HWnzBPn.js",
"https://cdn-client.medium.com/lite/assets/HorizontalScroller-DX9VVDGKwb.js",
"https://cdn-client.medium.com/lite/assets/BasicTooltip-BwWRxZW0H2.js",
"https://cdn-client.medium.com/lite/assets/TooltipPopover-BxIyQ3Eu21.js",
"https://cdn-client.medium.com/lite/assets/converters-CFjGMkrk4S.js",
"https://cdn-client.medium.com/lite/assets/PageTitle-x-SPpU_qxv.js",
"https://cdn-client.medium.com/lite/assets/MaxWidth-clK7fEc193.js",
"https://cdn-client.medium.com/lite/assets/useClientNavigation-BXw3WK3zM9.js",
"https://cdn-client.medium.com/lite/assets/FriendBadge-BK5Uzu_yjN.js",
"https://cdn-client.medium.com/lite/assets/useToggle-w-u1N56vw9.js",
"https://cdn-client.medium.com/lite/assets/pluralize-Dcs29zKoPV.js",
"https://cdn-client.medium.com/lite/assets/useTimer-ahkFCywLqc.js",
"https://cdn-client.medium.com/lite/assets/useShimmerAnimation-BcTM3YJxhP.js",
"https://cdn-client.medium.com/lite/assets/MembershipBadge-C58bOtXo9-.js",
"https://cdn-client.medium.com/lite/assets/BookVerificationBadge-BcsZNIdQVu.js",
"https://cdn-client.medium.com/lite/assets/index-C5HGGZaA1H.js",
"https://cdn-client.medium.com/lite/assets/helpers-DuUr-IXW9T.js",
"https://cdn-client.medium.com/lite/assets/formatMembershipDate-CKGLHYfZsK.js",
"https://cdn-client.medium.com/lite/assets/timeFormatUtils-g2m1dSWL0J.js",
"https://cdn-client.medium.com/lite/assets/BaseDialog-C9Jh0Y4bdK.js",
"https://cdn-client.medium.com/lite/assets/ContentShimmer-Bv6xn6myOY.js",
"https://cdn-client.medium.com/lite/assets/CreditCardIcon-BWlhbufPaa.js",
"https://cdn-client.medium.com/lite/assets/visa-C-R6AjgiLa.js",
"https://cdn-client.medium.com/lite/assets/memberships-DNbl3TjKMI.js",
"https://cdn-client.medium.com/lite/assets/format-ORhoBzAfcE.js",
"https://cdn-client.medium.com/lite/assets/useCurrentMembershipType-CAr67orssE.js",
"https://cdn-client.medium.com/lite/assets/index-C8Wk7uW_bo.js",
"https://cdn-client.medium.com/lite/assets/PresentationTracker-CPzAv_34HT.js",
"https://cdn-client.medium.com/lite/assets/IntersectionObserver-BIUJnjgxm4.js",
"https://cdn-client.medium.com/lite/assets/lists-filled-24px-DOUw0bo_Gh.js",
"https://cdn-client.medium.com/lite/assets/SearchInline-Chjfzh1JdW.js",
"https://cdn-client.medium.com/lite/assets/tinykeys.module-DmIGHpV-B2.js",
"https://cdn-client.medium.com/lite/assets/PopoverMenu-gHTDy3R03r.js",
"https://cdn-client.medium.com/lite/assets/SquareImage-BUQhjK_Bb3.js",
"https://cdn-client.medium.com/lite/assets/MiroImage-BKAA34AFMI.js",
"https://cdn-client.medium.com/lite/assets/topic-Dae92C93Dd.js",
"https://cdn-client.medium.com/lite/assets/HorizontalDivider-HaOMjvrYve.js",
"https://cdn-client.medium.com/lite/assets/explore-B7IXIWKhqc.js",
"https://cdn-client.medium.com/lite/assets/Date-BkxdzGCRUJ.js",
"https://cdn-client.medium.com/lite/assets/constant-BbXslOR1r4.js",
"https://cdn-client.medium.com/lite/assets/CollectionAvatar-BbsSaHGDVu.js",
"https://cdn-client.medium.com/lite/assets/publications-cDhzOwdN10.js",
"https://cdn-client.medium.com/lite/assets/users-DYlYbYHnyO.js",
"https://cdn-client.medium.com/lite/assets/isUserVerifiedBookAuthor-rzeUseXpbO.js",
"https://cdn-client.medium.com/lite/assets/Modal-DTdLYpDmnA.js",
"https://cdn-client.medium.com/lite/assets/FollowingBadge-PGSMEByPHA.js",
"https://cdn-client.medium.com/lite/assets/useUserViewerEdge-CfC4ZfuTBQ.js",
"https://cdn-client.medium.com/lite/assets/TopNav-AHlAsf9hna.js",
"https://cdn-client.medium.com/lite/assets/theme-D11jH8i4PC.js",
"https://cdn-client.medium.com/lite/assets/chevron-down-BNelhiVh37.js",
"https://cdn-client.medium.com/lite/assets/plus-16px-0xnU0BlbCH.js",
"https://cdn-client.medium.com/lite/assets/error-24px-BWFAb7mHfq.js",
"https://cdn-client.medium.com/lite/assets/events-DFdwDo_p8W.js",
"https://cdn-client.medium.com/lite/assets/useViewerQuery-CG78_WM4_E.js",
"https://cdn-client.medium.com/lite/assets/twitter-icon-Bzemimo3um.js",
"https://cdn-client.medium.com/lite/assets/Checkbox-CEpZybmOzs.js",
"https://cdn-client.medium.com/lite/assets/useServiceAuthUrl-CYlhWiPOAI.js",
"https://cdn-client.medium.com/lite/assets/facebook-icon-CTFtEBVMAi.js",
"https://cdn-client.medium.com/lite/assets/FloatingPortal-CIHz4Ddj70.js",
"https://cdn-client.medium.com/lite/assets/DefaultToastMessage-ClK0cOtZW0.js",
"https://cdn-client.medium.com/lite/assets/useSendVerificationEmail-D-T6JolvWw.js",
"https://cdn-client.medium.com/lite/assets/responsesSidebarUtils-P1lCDAk5tr.js",
"https://cdn-client.medium.com/lite/assets/useMutingActions-QBw4WbTfeu.js",
"https://cdn-client.medium.com/lite/assets/useCollectionViewerEdge-DSA5r45PFQ.js",
"https://cdn-client.medium.com/lite/assets/useTagViewerEdge-BPwelZwdZz.js",
"https://cdn-client.medium.com/lite/assets/cache-DAeQfZIjhg.js",
"https://cdn-client.medium.com/lite/assets/UndeletePostMutation-DqELsRmbR9.js",
"https://cdn-client.medium.com/lite/assets/DeletePostMutation-D_-pLOIrAI.js",
"https://cdn-client.medium.com/lite/assets/DeletePostMutation.generated.graphql-CxsaFDqjlm.js",
"https://cdn-client.medium.com/lite/assets/users-BgK6Dc8vo4.js",
"https://cdn-client.medium.com/lite/assets/useFollowCollection-CGhsA02QO9.js",
"https://cdn-client.medium.com/lite/assets/useSendFollowEvent-BMGhU8TxT5.js",
"https://cdn-client.medium.com/lite/assets/useFollowUser-BgNqlRDj47.js",
"https://cdn-client.medium.com/lite/assets/dom-DDy6wUDMDa.js",
"https://cdn-client.medium.com/lite/assets/MiddotContainer-FtEBnsxBEZ.js",
"https://cdn-client.medium.com/lite/assets/FollowSubscribeButton-Dooa_hj_C6.js",
"https://cdn-client.medium.com/lite/assets/chevron-down-24px-CBba64dKHK.js",
"https://cdn-client.medium.com/lite/assets/RadioGroup-DZhrVcchj7.js",
"https://cdn-client.medium.com/lite/assets/numbers-C1x1ytAIkn.js"
],
"css": []
},
"routes/pages/entity-routes/ShowEntitySlug": {
"id": "routes/pages/entity-routes/ShowEntitySlug",
"parentId": "root",
"path": ":slug",
"hasAction": false,
"hasLoader": true,
"hasClientAction": false,
"hasClientLoader": true,
"hasClientMiddleware": false,
"hasDefaultExport": true,
"hasErrorBoundary": false,
"module": "https://cdn-client.medium.com/lite/assets/ShowEntitySlug-DrZts1oV_Z.js",
"imports": [
"https://cdn-client.medium.com/lite/assets/vendor-react-CpuriD7yOR.js",
"https://cdn-client.medium.com/lite/assets/createEntityRoute-BJd2kwGj0Q.js",
"https://cdn-client.medium.com/lite/assets/entityPageRegistry-BzOK2jlsWT.js",
"https://cdn-client.medium.com/lite/assets/NotFoundHandler-D2FdvSX1Qb.js",
"https://cdn-client.medium.com/lite/assets/NotFoundScreen-Cbxs6enf11.js",
"https://cdn-client.medium.com/lite/assets/MiddotContainer-FtEBnsxBEZ.js",
"https://cdn-client.medium.com/lite/assets/RoutingContext-N-dlYRSyd6.js",
"https://cdn-client.medium.com/lite/assets/vendor-fela-Cy19ySZi4y.js",
"https://cdn-client.medium.com/lite/assets/vendor-lodash-C0lcM7XZQI.js",
"https://cdn-client.medium.com/lite/assets/vendor-apollo-B0FLDVoLJa.js",
"https://cdn-client.medium.com/lite/assets/Layer-BoAybuFPli.js",
"https://cdn-client.medium.com/lite/assets/PresentationTracker-CPzAv_34HT.js",
"https://cdn-client.medium.com/lite/assets/IntersectionObserver-BIUJnjgxm4.js",
"https://cdn-client.medium.com/lite/assets/BasicTooltip-BwWRxZW0H2.js",
"https://cdn-client.medium.com/lite/assets/TooltipPopover-BxIyQ3Eu21.js",
"https://cdn-client.medium.com/lite/assets/Popover-BEHTzJA9dJ.js",
"https://cdn-client.medium.com/lite/assets/index.esm-dkQXnoW74C.js",
"https://cdn-client.medium.com/lite/assets/events-DFdwDo_p8W.js",
"https://cdn-client.medium.com/lite/assets/dom-DDy6wUDMDa.js",
"https://cdn-client.medium.com/lite/assets/converters-CFjGMkrk4S.js",
"https://cdn-client.medium.com/lite/assets/CircularImage-ChC4HYcXtN.js",
"https://cdn-client.medium.com/lite/assets/MiroImage-BKAA34AFMI.js",
"https://cdn-client.medium.com/lite/assets/miroImage-DnVurRI6Te.js",
"https://cdn-client.medium.com/lite/assets/ZoomableImage-CMLNdmn9hp.js",
"https://cdn-client.medium.com/lite/assets/Dialog-AkVTR0-Mir.js",
"https://cdn-client.medium.com/lite/assets/CloseButton-C6-EVtDglw.js",
"https://cdn-client.medium.com/lite/assets/pluralize-Dcs29zKoPV.js",
"https://cdn-client.medium.com/lite/assets/useReduxSusi-BDnRTMQqz7.js",
"https://cdn-client.medium.com/lite/assets/actions-BBSCsZ7iAN.js",
"https://cdn-client.medium.com/lite/assets/FollowSubscribeButton-Dooa_hj_C6.js",
"https://cdn-client.medium.com/lite/assets/useToggle-w-u1N56vw9.js",
"https://cdn-client.medium.com/lite/assets/chevron-down-24px-CBba64dKHK.js",
"https://cdn-client.medium.com/lite/assets/RadioGroup-DZhrVcchj7.js",
"https://cdn-client.medium.com/lite/assets/Modal-DTdLYpDmnA.js",
"https://cdn-client.medium.com/lite/assets/useClientResponsiveSize-DMj2FEayo4.js",
"https://cdn-client.medium.com/lite/assets/PopoverMenu-gHTDy3R03r.js",
"https://cdn-client.medium.com/lite/assets/useFollowUser-BgNqlRDj47.js",
"https://cdn-client.medium.com/lite/assets/useClientViewerOrVisitorId-CBF3XYp5m3.js",
"https://cdn-client.medium.com/lite/assets/cache-DAeQfZIjhg.js",
"https://cdn-client.medium.com/lite/assets/useUserViewerEdge-CfC4ZfuTBQ.js",
"https://cdn-client.medium.com/lite/assets/useSendFollowEvent-BMGhU8TxT5.js",
"https://cdn-client.medium.com/lite/assets/selectors-Civm4lBZJ3.js",
"https://cdn-client.medium.com/lite/assets/users-DYlYbYHnyO.js",
"https://cdn-client.medium.com/lite/assets/MaxWidth-clK7fEc193.js",
"https://cdn-client.medium.com/lite/assets/ErrorScreen-DfI8g0yVkR.js",
"https://cdn-client.medium.com/lite/assets/BookVerificationBadge-BcsZNIdQVu.js",
"https://cdn-client.medium.com/lite/assets/useShimmerAnimation-BcTM3YJxhP.js",
"https://cdn-client.medium.com/lite/assets/isUserVerifiedBookAuthor-rzeUseXpbO.js",
"https://cdn-client.medium.com/lite/assets/numbers-C1x1ytAIkn.js",
"https://cdn-client.medium.com/lite/assets/FriendBadge-BK5Uzu_yjN.js",
"https://cdn-client.medium.com/lite/assets/Date-BkxdzGCRUJ.js",
"https://cdn-client.medium.com/lite/assets/index-C5HGGZaA1H.js",
"https://cdn-client.medium.com/lite/assets/index-9bMYTUUBe-.js",
"https://cdn-client.medium.com/lite/assets/GlobalNavigationContext-CFsD4sSJxy.js",
"https://cdn-client.medium.com/lite/assets/useIsomorphicLayoutEffect-mTFvRtpWc0.js",
"https://cdn-client.medium.com/lite/assets/OpenInApp-DfYOBi2m2t.js",
"https://cdn-client.medium.com/lite/assets/useragent-DM_y0Y6BQs.js",
"https://cdn-client.medium.com/lite/assets/SusiClickable-D3bZIDOhCt.js",
"https://cdn-client.medium.com/lite/assets/UserIdStorage-CucM5frZ7R.js",
"https://cdn-client.medium.com/lite/assets/useGoogleTagEvent-BVvaY-rQcL.js",
"https://cdn-client.medium.com/lite/assets/SettingsPage-1nQFOpO_S1.js",
"https://cdn-client.medium.com/lite/assets/PageHeader-vZiraAAZX8.js",
"https://cdn-client.medium.com/lite/assets/PageNavigationBar-BE30YyYyoY.js",
"https://cdn-client.medium.com/lite/assets/ResizeObserver-R46HWnzBPn.js",
"https://cdn-client.medium.com/lite/assets/HorizontalScroller-DX9VVDGKwb.js",
"https://cdn-client.medium.com/lite/assets/color-DcO1XOeLw1.js",
"https://cdn-client.medium.com/lite/assets/PageTitle-x-SPpU_qxv.js",
"https://cdn-client.medium.com/lite/assets/useClientNavigation-BXw3WK3zM9.js",
"https://cdn-client.medium.com/lite/assets/visits-BnXWJEFetB.js",
"https://cdn-client.medium.com/lite/assets/useTimer-ahkFCywLqc.js",
"https://cdn-client.medium.com/lite/assets/MembershipBadge-C58bOtXo9-.js",
"https://cdn-client.medium.com/lite/assets/LoadingSpinner-C-Fc1l7DXa.js",
"https://cdn-client.medium.com/lite/assets/helpers-DuUr-IXW9T.js",
"https://cdn-client.medium.com/lite/assets/formatMembershipDate-CKGLHYfZsK.js",
"https://cdn-client.medium.com/lite/assets/timeFormatUtils-g2m1dSWL0J.js",
"https://cdn-client.medium.com/lite/assets/BaseDialog-C9Jh0Y4bdK.js",
"https://cdn-client.medium.com/lite/assets/ContentShimmer-Bv6xn6myOY.js",
"https://cdn-client.medium.com/lite/assets/CreditCardIcon-BWlhbufPaa.js",
"https://cdn-client.medium.com/lite/assets/visa-C-R6AjgiLa.js",
"https://cdn-client.medium.com/lite/assets/memberships-DNbl3TjKMI.js",
"https://cdn-client.medium.com/lite/assets/format-ORhoBzAfcE.js",
"https://cdn-client.medium.com/lite/assets/useCurrentMembershipType-CAr67orssE.js",
"https://cdn-client.medium.com/lite/assets/index-C8Wk7uW_bo.js",
"https://cdn-client.medium.com/lite/assets/lists-filled-24px-DOUw0bo_Gh.js",
"https://cdn-client.medium.com/lite/assets/SearchInline-Chjfzh1JdW.js",
"https://cdn-client.medium.com/lite/assets/tinykeys.module-DmIGHpV-B2.js",
"https://cdn-client.medium.com/lite/assets/SquareImage-BUQhjK_Bb3.js",
"https://cdn-client.medium.com/lite/assets/topic-Dae92C93Dd.js",
"https://cdn-client.medium.com/lite/assets/actions-BuaTAfK8Ws.js",
"https://cdn-client.medium.com/lite/assets/HorizontalDivider-HaOMjvrYve.js",
"https://cdn-client.medium.com/lite/assets/explore-B7IXIWKhqc.js",
"https://cdn-client.medium.com/lite/assets/constant-BbXslOR1r4.js",
"https://cdn-client.medium.com/lite/assets/CollectionAvatar-BbsSaHGDVu.js",
"https://cdn-client.medium.com/lite/assets/publications-cDhzOwdN10.js",
"https://cdn-client.medium.com/lite/assets/Logo-CAikMcA5xF.js",
"https://cdn-client.medium.com/lite/assets/FollowingBadge-PGSMEByPHA.js",
"https://cdn-client.medium.com/lite/assets/usePreference-D22I6RuVgo.js",
"https://cdn-client.medium.com/lite/assets/TopNav-AHlAsf9hna.js",
"https://cdn-client.medium.com/lite/assets/theme-D11jH8i4PC.js",
"https://cdn-client.medium.com/lite/assets/chevron-down-BNelhiVh37.js",
"https://cdn-client.medium.com/lite/assets/plus-16px-0xnU0BlbCH.js",
"https://cdn-client.medium.com/lite/assets/error-24px-BWFAb7mHfq.js",
"https://cdn-client.medium.com/lite/assets/useViewerQuery-CG78_WM4_E.js",
"https://cdn-client.medium.com/lite/assets/context-Dpzg9QPdZT.js"
],
"css": []
},
"routes/pages/entity-routes/ShowMediumRoot": {
"id": "routes/pages/entity-routes/ShowMediumRoot",
"parentId": "root",
"index": true,
"hasAction": false,
"hasLoader": true,
"hasClientAction": false,
"hasClientLoader": true,
"hasClientMiddleware": false,
"hasDefaultExport": true,
"hasErrorBoundary": false,
"module": "https://cdn-client.medium.com/lite/assets/ShowMediumRoot-Ck37vLPOIu.js",
"imports": [
"https://cdn-client.medium.com/lite/assets/vendor-react-CpuriD7yOR.js",
"https://cdn-client.medium.com/lite/assets/createEntityRoute-BJd2kwGj0Q.js",
"https://cdn-client.medium.com/lite/assets/entityPageRegistry-BzOK2jlsWT.js",
"https://cdn-client.medium.com/lite/assets/NotFoundHandler-D2FdvSX1Qb.js",
"https://cdn-client.medium.com/lite/assets/NotFoundScreen-Cbxs6enf11.js",
"https://cdn-client.medium.com/lite/assets/MiddotContainer-FtEBnsxBEZ.js",
"https://cdn-client.medium.com/lite/assets/RoutingContext-N-dlYRSyd6.js",
"https://cdn-client.medium.com/lite/assets/vendor-fela-Cy19ySZi4y.js",
"https://cdn-client.medium.com/lite/assets/vendor-lodash-C0lcM7XZQI.js",
"https://cdn-client.medium.com/lite/assets/vendor-apollo-B0FLDVoLJa.js",
"https://cdn-client.medium.com/lite/assets/Layer-BoAybuFPli.js",
"https://cdn-client.medium.com/lite/assets/PresentationTracker-CPzAv_34HT.js",
"https://cdn-client.medium.com/lite/assets/IntersectionObserver-BIUJnjgxm4.js",
"https://cdn-client.medium.com/lite/assets/BasicTooltip-BwWRxZW0H2.js",
"https://cdn-client.medium.com/lite/assets/TooltipPopover-BxIyQ3Eu21.js",
"https://cdn-client.medium.com/lite/assets/Popover-BEHTzJA9dJ.js",
"https://cdn-client.medium.com/lite/assets/index.esm-dkQXnoW74C.js",
"https://cdn-client.medium.com/lite/assets/events-DFdwDo_p8W.js",
"https://cdn-client.medium.com/lite/assets/dom-DDy6wUDMDa.js",
"https://cdn-client.medium.com/lite/assets/converters-CFjGMkrk4S.js",
"https://cdn-client.medium.com/lite/assets/CircularImage-ChC4HYcXtN.js",
"https://cdn-client.medium.com/lite/assets/MiroImage-BKAA34AFMI.js",
"https://cdn-client.medium.com/lite/assets/miroImage-DnVurRI6Te.js",
"https://cdn-client.medium.com/lite/assets/ZoomableImage-CMLNdmn9hp.js",
"https://cdn-client.medium.com/lite/assets/Dialog-AkVTR0-Mir.js",
"https://cdn-client.medium.com/lite/assets/CloseButton-C6-EVtDglw.js",
"https://cdn-client.medium.com/lite/assets/pluralize-Dcs29zKoPV.js",
"https://cdn-client.medium.com/lite/assets/useReduxSusi-BDnRTMQqz7.js",
"https://cdn-client.medium.com/lite/assets/actions-BBSCsZ7iAN.js",
"https://cdn-client.medium.com/lite/assets/FollowSubscribeButton-Dooa_hj_C6.js",
"https://cdn-client.medium.com/lite/assets/useToggle-w-u1N56vw9.js",
"https://cdn-client.medium.com/lite/assets/chevron-down-24px-CBba64dKHK.js",
"https://cdn-client.medium.com/lite/assets/RadioGroup-DZhrVcchj7.js",
"https://cdn-client.medium.com/lite/assets/Modal-DTdLYpDmnA.js",
"https://cdn-client.medium.com/lite/assets/useClientResponsiveSize-DMj2FEayo4.js",
"https://cdn-client.medium.com/lite/assets/PopoverMenu-gHTDy3R03r.js",
"https://cdn-client.medium.com/lite/assets/useFollowUser-BgNqlRDj47.js",
"https://cdn-client.medium.com/lite/assets/useClientViewerOrVisitorId-CBF3XYp5m3.js",
"https://cdn-client.medium.com/lite/assets/cache-DAeQfZIjhg.js",
"https://cdn-client.medium.com/lite/assets/useUserViewerEdge-CfC4ZfuTBQ.js",
"https://cdn-client.medium.com/lite/assets/useSendFollowEvent-BMGhU8TxT5.js",
"https://cdn-client.medium.com/lite/assets/selectors-Civm4lBZJ3.js",
"https://cdn-client.medium.com/lite/assets/users-DYlYbYHnyO.js",
"https://cdn-client.medium.com/lite/assets/MaxWidth-clK7fEc193.js",
"https://cdn-client.medium.com/lite/assets/ErrorScreen-DfI8g0yVkR.js",
"https://cdn-client.medium.com/lite/assets/BookVerificationBadge-BcsZNIdQVu.js",
"https://cdn-client.medium.com/lite/assets/useShimmerAnimation-BcTM3YJxhP.js",
"https://cdn-client.medium.com/lite/assets/isUserVerifiedBookAuthor-rzeUseXpbO.js",
"https://cdn-client.medium.com/lite/assets/numbers-C1x1ytAIkn.js",
"https://cdn-client.medium.com/lite/assets/FriendBadge-BK5Uzu_yjN.js",
"https://cdn-client.medium.com/lite/assets/Date-BkxdzGCRUJ.js",
"https://cdn-client.medium.com/lite/assets/index-C5HGGZaA1H.js",
"https://cdn-client.medium.com/lite/assets/index-9bMYTUUBe-.js",
"https://cdn-client.medium.com/lite/assets/GlobalNavigationContext-CFsD4sSJxy.js",
"https://cdn-client.medium.com/lite/assets/useIsomorphicLayoutEffect-mTFvRtpWc0.js",
"https://cdn-client.medium.com/lite/assets/OpenInApp-DfYOBi2m2t.js",
"https://cdn-client.medium.com/lite/assets/useragent-DM_y0Y6BQs.js",
"https://cdn-client.medium.com/lite/assets/SusiClickable-D3bZIDOhCt.js",
"https://cdn-client.medium.com/lite/assets/UserIdStorage-CucM5frZ7R.js",
"https://cdn-client.medium.com/lite/assets/useGoogleTagEvent-BVvaY-rQcL.js",
"https://cdn-client.medium.com/lite/assets/SettingsPage-1nQFOpO_S1.js",
"https://cdn-client.medium.com/lite/assets/PageHeader-vZiraAAZX8.js",
"https://cdn-client.medium.com/lite/assets/PageNavigationBar-BE30YyYyoY.js",
"https://cdn-client.medium.com/lite/assets/ResizeObserver-R46HWnzBPn.js",
"https://cdn-client.medium.com/lite/assets/HorizontalScroller-DX9VVDGKwb.js",
"https://cdn-client.medium.com/lite/assets/color-DcO1XOeLw1.js",
"https://cdn-client.medium.com/lite/assets/PageTitle-x-SPpU_qxv.js",
"https://cdn-client.medium.com/lite/assets/useClientNavigation-BXw3WK3zM9.js",
"https://cdn-client.medium.com/lite/assets/visits-BnXWJEFetB.js",
"https://cdn-client.medium.com/lite/assets/useTimer-ahkFCywLqc.js",
"https://cdn-client.medium.com/lite/assets/MembershipBadge-C58bOtXo9-.js",
"https://cdn-client.medium.com/lite/assets/LoadingSpinner-C-Fc1l7DXa.js",
"https://cdn-client.medium.com/lite/assets/helpers-DuUr-IXW9T.js",
"https://cdn-client.medium.com/lite/assets/formatMembershipDate-CKGLHYfZsK.js",
"https://cdn-client.medium.com/lite/assets/timeFormatUtils-g2m1dSWL0J.js",
"https://cdn-client.medium.com/lite/assets/BaseDialog-C9Jh0Y4bdK.js",
"https://cdn-client.medium.com/lite/assets/ContentShimmer-Bv6xn6myOY.js",
"https://cdn-client.medium.com/lite/assets/CreditCardIcon-BWlhbufPaa.js",
"https://cdn-client.medium.com/lite/assets/visa-C-R6AjgiLa.js",
"https://cdn-client.medium.com/lite/assets/memberships-DNbl3TjKMI.js",
"https://cdn-client.medium.com/lite/assets/format-ORhoBzAfcE.js",
"https://cdn-client.medium.com/lite/assets/useCurrentMembershipType-CAr67orssE.js",
"https://cdn-client.medium.com/lite/assets/index-C8Wk7uW_bo.js",
"https://cdn-client.medium.com/lite/assets/lists-filled-24px-DOUw0bo_Gh.js",
"https://cdn-client.medium.com/lite/assets/SearchInline-Chjfzh1JdW.js",
"https://cdn-client.medium.com/lite/assets/tinykeys.module-DmIGHpV-B2.js",
"https://cdn-client.medium.com/lite/assets/SquareImage-BUQhjK_Bb3.js",
"https://cdn-client.medium.com/lite/assets/topic-Dae92C93Dd.js",
"https://cdn-client.medium.com/lite/assets/actions-BuaTAfK8Ws.js",
"https://cdn-client.medium.com/lite/assets/HorizontalDivider-HaOMjvrYve.js",
"https://cdn-client.medium.com/lite/assets/explore-B7IXIWKhqc.js",
"https://cdn-client.medium.com/lite/assets/constant-BbXslOR1r4.js",
"https://cdn-client.medium.com/lite/assets/CollectionAvatar-BbsSaHGDVu.js",
"https://cdn-client.medium.com/lite/assets/publications-cDhzOwdN10.js",
"https://cdn-client.medium.com/lite/assets/Logo-CAikMcA5xF.js",
"https://cdn-client.medium.com/lite/assets/FollowingBadge-PGSMEByPHA.js",
"https://cdn-client.medium.com/lite/assets/usePreference-D22I6RuVgo.js",
"https://cdn-client.medium.com/lite/assets/TopNav-AHlAsf9hna.js",
"https://cdn-client.medium.com/lite/assets/theme-D11jH8i4PC.js",
"https://cdn-client.medium.com/lite/assets/chevron-down-BNelhiVh37.js",
"https://cdn-client.medium.com/lite/assets/plus-16px-0xnU0BlbCH.js",
"https://cdn-client.medium.com/lite/assets/error-24px-BWFAb7mHfq.js",
"https://cdn-client.medium.com/lite/assets/useViewerQuery-CG78_WM4_E.js",
"https://cdn-client.medium.com/lite/assets/context-Dpzg9QPdZT.js"
],
"css": []
}
},
"url": "https://cdn-client.medium.com/lite/static/manifest-52af2e45.js",
"version": "52af2e45"
};
window.__reactRouterRouteModules = {"root":route0,"routes/pages/entity-routes/ShowEntitySlug":route1};
import("https://cdn-client.medium.com/lite/assets/entry.client-DtJLrZR8VY.js");</script><!--$--><script>window.__reactRouterContext.streamController.enqueue("[{\"_1\":2,\"_3\":-5,\"_4\":-5},\"loaderData\",{\"_5\":6},\"actionData\",\"errors\",\"routes/pages/entity-routes/ShowEntitySlug\",{}]\n");</script><!--$--><script>window.__reactRouterContext.streamController.close();</script><!--/$--><!--/$--><script>window.__APOLLO_STATE__ = {"ROOT_QUERY":{"__typename":"Query","viewer":null,"postResult({\"id\":\"a64152b37c35\"})":{"__ref":"Post:a64152b37c35"}},"LinkedAccounts:ac9d9a35533e":{"__typename":"LinkedAccounts","mastodon":null,"id":"ac9d9a35533e"},"NewsletterV3:e877d56ca397":{"__typename":"NewsletterV3","id":"e877d56ca397","type":"NEWSLETTER_TYPE_AUTHOR","slug":"ac9d9a35533e","name":"ac9d9a35533e","collection":null,"user":{"__ref":"User:ac9d9a35533e"}},"User:ac9d9a35533e":{"__typename":"User","id":"ac9d9a35533e","name":"Andrej Karpathy","username":"karpathy","newsletterV3":{"__ref":"NewsletterV3:e877d56ca397"},"linkedAccounts":{"__ref":"LinkedAccounts:ac9d9a35533e"},"isSuspended":false,"customDomainState":{"__typename":"CustomDomainState","live":{"__typename":"CustomDomain","domain":"karpathy.medium.com"}},"hasSubdomain":true,"verifications":{"__typename":"VerifiedInfo","isBookAuthor":false},"imageId":"0*8ldFdx9B6FhSkQmV.jpeg","socialStats":{"__typename":"SocialStats","followerCount":61372,"followingCount":183,"collectionFollowingCount":3},"bio":"I like to train deep neural nets on large datasets.","allowNotes":true,"membership":null,"viewerEdge":{"__ref":"UserViewerEdge:userId:ac9d9a35533e-viewerId:lo_93d1296512d8"},"twitterScreenName":"karpathy"},"Topic:decb52b64abf":{"__typename":"Topic","slug":"programming","id":"decb52b64abf"},"Paragraph:b1853291a8ec_0":{"__typename":"Paragraph","id":"b1853291a8ec_0","name":"62ab","type":"H3","href":null,"layout":null,"metadata":null,"text":"Software 2.0","hasDropCap":null,"dropCapImage":null,"markups":[],"codeBlockMetadata":null,"iframe":null,"mixtapeMetadata":null},"Paragraph:b1853291a8ec_1":{"__typename":"Paragraph","id":"b1853291a8ec_1","name":"de15","type":"P","href":null,"layout":null,"metadata":null,"text":"I sometimes see people refer to neural networks as just “another tool in your machine learning toolbox”. They have some pros and cons, they work here or there, and sometimes you can use them to win Kaggle competitions. Unfortunately, this interpretation completely misses the forest for the trees. Neural networks are not just another classifier, they represent the beginning of a fundamental shift in how we develop software. They are Software 2.0.","hasDropCap":null,"dropCapImage":null,"markups":[],"codeBlockMetadata":null,"iframe":null,"mixtapeMetadata":null},"Paragraph:b1853291a8ec_2":{"__typename":"Paragraph","id":"b1853291a8ec_2","name":"2814","type":"P","href":null,"layout":null,"metadata":null,"text":"The “classical stack” of Software 1.0 is what we’re all familiar with — it is written in languages such as Python, C++, etc. It consists of explicit instructions to the computer written by a programmer. By writing each line of code, the programmer identifies a specific point in program space with some desirable behavior.","hasDropCap":null,"dropCapImage":null,"markups":[{"__typename":"Markup","type":"STRONG","start":25,"end":37,"href":null,"anchorType":null,"userId":null,"linkMetadata":null}],"codeBlockMetadata":null,"iframe":null,"mixtapeMetadata":null},"Paragraph:b1853291a8ec_3":{"__typename":"Paragraph","id":"b1853291a8ec_3","name":"e3a8","type":"IMG","href":null,"layout":"INSET_CENTER","metadata":{"__typename":"ImageMetadata","id":"1*CHcu2L0NmAZwCpQgmS1ByA.jpeg","originalHeight":234,"originalWidth":1017,"focusPercentX":null,"focusPercentY":null,"alt":null},"text":"","hasDropCap":null,"dropCapImage":null,"markups":[],"codeBlockMetadata":null,"iframe":null,"mixtapeMetadata":null},"Paragraph:b1853291a8ec_4":{"__typename":"Paragraph","id":"b1853291a8ec_4","name":"b05f","type":"P","href":null,"layout":null,"metadata":null,"text":"In contrast, Software 2.0 is written in much more abstract, human unfriendly language, such as the weights of a neural network. No human is involved in writing this code because there are a lot of weights (typical networks might have millions), and coding directly in weights is kind of hard (I tried).","hasDropCap":null,"dropCapImage":null,"markups":[{"__typename":"Markup","type":"STRONG","start":13,"end":25,"href":null,"anchorType":null,"userId":null,"linkMetadata":null}],"codeBlockMetadata":null,"iframe":null,"mixtapeMetadata":null},"Paragraph:b1853291a8ec_5":{"__typename":"Paragraph","id":"b1853291a8ec_5","name":"4402","type":"IMG","href":null,"layout":"INSET_CENTER","metadata":{"__typename":"ImageMetadata","id":"1*6EB1Xue1wM_QP0IIzXphQA.png","originalHeight":528,"originalWidth":850,"focusPercentX":null,"focusPercentY":null,"alt":null},"text":"","hasDropCap":null,"dropCapImage":null,"markups":[],"codeBlockMetadata":null,"iframe":null,"mixtapeMetadata":null},"Paragraph:b1853291a8ec_6":{"__typename":"Paragraph","id":"b1853291a8ec_6","name":"eabb","type":"P","href":null,"layout":null,"metadata":null,"text":"Instead, our approach is to specify some goal on the behavior of a desirable program (e.g., “satisfy a dataset of input output pairs of examples”, or “win a game of Go”), write a rough skeleton of the code (i.e. a neural net architecture) that identifies a subset of program space to search, and use the computational resources at our disposal to search this space for a program that works. In the case of neural networks, we restrict the search to a continuous subset of the program space where the search process can be made (somewhat surprisingly) efficient with backpropagation and stochastic gradient descent.","hasDropCap":null,"dropCapImage":null,"markups":[],"codeBlockMetadata":null,"iframe":null,"mixtapeMetadata":null},"Paragraph:b1853291a8ec_7":{"__typename":"Paragraph","id":"b1853291a8ec_7","name":"2c5c","type":"IMG","href":null,"layout":"INSET_CENTER","metadata":{"__typename":"ImageMetadata","id":"1*5NG3U8MsaTqmQpjkr_-UOw.png","originalHeight":700,"originalWidth":1219,"focusPercentX":null,"focusPercentY":null,"alt":null},"text":"","hasDropCap":null,"dropCapImage":null,"markups":[],"codeBlockMetadata":null,"iframe":null,"mixtapeMetadata":null},"Paragraph:b1853291a8ec_8":{"__typename":"Paragraph","id":"b1853291a8ec_8","name":"26d1","type":"P","href":null,"layout":null,"metadata":null,"text":"To make the analogy explicit, in Software 1.0, human-engineered source code (e.g. some .cpp files) is compiled into a binary that does useful work. In Software 2.0 most often the source code comprises 1) the dataset that defines the desirable behavior and 2) the neural net architecture that gives the rough skeleton of the code, but with many details (the weights) to be filled in. The process of training the neural network compiles the dataset into the binary — the final neural network. In most practical applications today, the neural net architectures and the training systems are increasingly standardized into a commodity, so most of the active “software development” takes the form of curating, growing, massaging and cleaning labeled datasets. This is fundamentally altering the programming paradigm by which we iterate on our software, as the teams split in two: the 2.0 programmers (data labelers) edit and grow the datasets, while a few 1.0 programmers maintain and iterate on the surrounding training code infrastructure, analytics, visualizations and labeling interfaces.","hasDropCap":null,"dropCapImage":null,"markups":[],"codeBlockMetadata":null,"iframe":null,"mixtapeMetadata":null},"Paragraph:b1853291a8ec_9":{"__typename":"Paragraph","id":"b1853291a8ec_9","name":"d3ca","type":"P","href":null,"layout":null,"metadata":null,"text":"It turns out that a large portion of real-world problems have the property that it is significantly easier to collect the data (or more generally, identify a desirable behavior) than to explicitly write the program. Because of this and many other benefits of Software 2.0 programs that I will go into below, we are witnessing a massive transition across the industry where of a lot of 1.0 code is being ported into 2.0 code. Software (1.0) is eating the world, and now AI (Software 2.0) is eating software.","hasDropCap":null,"dropCapImage":null,"markups":[],"codeBlockMetadata":null,"iframe":null,"mixtapeMetadata":null},"Paragraph:b1853291a8ec_10":{"__typename":"Paragraph","id":"b1853291a8ec_10","name":"196c","type":"H3","href":null,"layout":null,"metadata":null,"text":"Ongoing transition","hasDropCap":null,"dropCapImage":null,"markups":[],"codeBlockMetadata":null,"iframe":null,"mixtapeMetadata":null},"Paragraph:b1853291a8ec_11":{"__typename":"Paragraph","id":"b1853291a8ec_11","name":"f0ad","type":"P","href":null,"layout":null,"metadata":null,"text":"Let’s briefly examine some concrete examples of this ongoing transition. In each of these areas we’ve seen improvements over the last few years when we give up on trying to address a complex problem by writing explicit code and instead transition the code into the 2.0 stack.","hasDropCap":null,"dropCapImage":null,"markups":[],"codeBlockMetadata":null,"iframe":null,"mixtapeMetadata":null},"Paragraph:b1853291a8ec_12":{"__typename":"Paragraph","id":"b1853291a8ec_12","name":"ebff","type":"P","href":null,"layout":null,"metadata":null,"text":"Visual Recognition used to consist of engineered features with a bit of machine learning sprinkled on top at the end (e.g., an SVM). Since then, we discovered much more powerful visual features by obtaining large datasets (e.g. ImageNet) and searching in the space of Convolutional Neural Network architectures. More recently, we don’t even trust ourselves to hand-code the architectures and we’ve begun searching over those as well.","hasDropCap":null,"dropCapImage":null,"markups":[{"__typename":"Markup","type":"A","start":404,"end":424,"href":"https:\u002F\u002Farxiv.org\u002Fabs\u002F1703.01041","anchorType":"LINK","userId":null,"linkMetadata":null},{"__typename":"Markup","type":"STRONG","start":0,"end":18,"href":null,"anchorType":null,"userId":null,"linkMetadata":null}],"codeBlockMetadata":null,"iframe":null,"mixtapeMetadata":null},"Paragraph:b1853291a8ec_13":{"__typename":"Paragraph","id":"b1853291a8ec_13","name":"6785","type":"P","href":null,"layout":null,"metadata":null,"text":"Speech recognition used to involve a lot of preprocessing, gaussian mixture models and hidden markov models, but today consist almost entirely of neural net stuff. A very related, often cited humorous quote attributed to Fred Jelinek from 1985 reads “Every time I fire a linguist, the performance of our speech recognition system goes up”.","hasDropCap":null,"dropCapImage":null,"markups":[{"__typename":"Markup","type":"A","start":113,"end":119,"href":"https:\u002F\u002Fgithub.com\u002Fsyhw\u002Fwer_are_we","anchorType":"LINK","userId":null,"linkMetadata":null},{"__typename":"Markup","type":"STRONG","start":0,"end":18,"href":null,"anchorType":null,"userId":null,"linkMetadata":null}],"codeBlockMetadata":null,"iframe":null,"mixtapeMetadata":null},"Paragraph:b1853291a8ec_14":{"__typename":"Paragraph","id":"b1853291a8ec_14","name":"b24e","type":"P","href":null,"layout":null,"metadata":null,"text":"Speech synthesis has historically been approached with various stitching mechanisms, but today the state of the art models are large ConvNets (e.g. WaveNet) that produce raw audio signal outputs.","hasDropCap":null,"dropCapImage":null,"markups":[{"__typename":"Markup","type":"A","start":148,"end":155,"href":"https:\u002F\u002Fdeepmind.com\u002Fblog\u002Fwavenet-launches-google-assistant\u002F","anchorType":"LINK","userId":null,"linkMetadata":null},{"__typename":"Markup","type":"STRONG","start":0,"end":16,"href":null,"anchorType":null,"userId":null,"linkMetadata":null}],"codeBlockMetadata":null,"iframe":null,"mixtapeMetadata":null},"Paragraph:b1853291a8ec_15":{"__typename":"Paragraph","id":"b1853291a8ec_15","name":"a4e3","type":"P","href":null,"layout":null,"metadata":null,"text":"Machine Translation has usually been approaches with phrase-based statistical techniques, but neural networks are quickly becoming dominant. My favorite architectures are trained in the multilingual setting, where a single model translates from any source language to any target language, and in weakly supervised (or entirely unsupervised) settings.","hasDropCap":null,"dropCapImage":null,"markups":[{"__typename":"Markup","type":"A","start":186,"end":206,"href":"https:\u002F\u002Farxiv.org\u002Fabs\u002F1611.04558","anchorType":"LINK","userId":null,"linkMetadata":null},{"__typename":"Markup","type":"A","start":327,"end":339,"href":"https:\u002F\u002Farxiv.org\u002Fabs\u002F1710.11041","anchorType":"LINK","userId":null,"linkMetadata":null},{"__typename":"Markup","type":"STRONG","start":0,"end":19,"href":null,"anchorType":null,"userId":null,"linkMetadata":null}],"codeBlockMetadata":null,"iframe":null,"mixtapeMetadata":null},"Paragraph:b1853291a8ec_16":{"__typename":"Paragraph","id":"b1853291a8ec_16","name":"386f","type":"P","href":null,"layout":null,"metadata":null,"text":"Games. Explicitly hand-coded Go playing programs have been developed for a long while, but AlphaGo Zero (a ConvNet that looks at the raw state of the board and plays a move) has now become by far the strongest player of the game. I expect we’re going to see very similar results in other areas, e.g. DOTA 2, or StarCraft.","hasDropCap":null,"dropCapImage":null,"markups":[{"__typename":"Markup","type":"A","start":91,"end":103,"href":"https:\u002F\u002Fdeepmind.com\u002Fblog\u002Falphago-zero-learning-scratch\u002F","anchorType":"LINK","userId":null,"linkMetadata":null},{"__typename":"Markup","type":"A","start":300,"end":306,"href":"https:\u002F\u002Fblog.openai.com\u002Fmore-on-dota-2\u002F","anchorType":"LINK","userId":null,"linkMetadata":null},{"__typename":"Markup","type":"A","start":311,"end":320,"href":"https:\u002F\u002Fdeepmind.com\u002Fblog\u002Fdeepmind-and-blizzard-open-starcraft-ii-ai-research-environment\u002F","anchorType":"LINK","userId":null,"linkMetadata":null},{"__typename":"Markup","type":"STRONG","start":0,"end":7,"href":null,"anchorType":null,"userId":null,"linkMetadata":null}],"codeBlockMetadata":null,"iframe":null,"mixtapeMetadata":null},"Paragraph:b1853291a8ec_17":{"__typename":"Paragraph","id":"b1853291a8ec_17","name":"089d","type":"P","href":null,"layout":null,"metadata":null,"text":"Databases. More traditional systems outside of Artificial Intelligence are also seeing early hints of a transition. For instance, “The Case for Learned Index Structures” replaces core components of a data management system with a neural network, outperforming cache-optimized B-Trees by up to 70% in speed while saving an order-of-magnitude in memory.","hasDropCap":null,"dropCapImage":null,"markups":[{"__typename":"Markup","type":"A","start":131,"end":168,"href":"https:\u002F\u002Farxiv.org\u002Fabs\u002F1712.01208","anchorType":"LINK","userId":null,"linkMetadata":null},{"__typename":"Markup","type":"STRONG","start":0,"end":9,"href":null,"anchorType":null,"userId":null,"linkMetadata":null}],"codeBlockMetadata":null,"iframe":null,"mixtapeMetadata":null},"Paragraph:b1853291a8ec_18":{"__typename":"Paragraph","id":"b1853291a8ec_18","name":"6c91","type":"P","href":null,"layout":null,"metadata":null,"text":"You’ll notice that many of my links above involve work done at Google. This is because Google is currently at the forefront of re-writing large chunks of itself into Software 2.0 code. “One model to rule them all” provides an early sketch of what this might look like, where the statistical strength of the individual domains is amalgamated into one consistent understanding of the world.","hasDropCap":null,"dropCapImage":null,"markups":[{"__typename":"Markup","type":"A","start":186,"end":212,"href":"https:\u002F\u002Farxiv.org\u002Fabs\u002F1706.05137","anchorType":"LINK","userId":null,"linkMetadata":null}],"codeBlockMetadata":null,"iframe":null,"mixtapeMetadata":null},"Paragraph:b1853291a8ec_19":{"__typename":"Paragraph","id":"b1853291a8ec_19","name":"0fac","type":"H3","href":null,"layout":null,"metadata":null,"text":"The benefits of Software 2.0","hasDropCap":null,"dropCapImage":null,"markups":[],"codeBlockMetadata":null,"iframe":null,"mixtapeMetadata":null},"Paragraph:b1853291a8ec_20":{"__typename":"Paragraph","id":"b1853291a8ec_20","name":"d6c4","type":"P","href":null,"layout":null,"metadata":null,"text":"Why should we prefer to port complex programs into Software 2.0? Clearly, one easy answer is that they work better in practice. However, there are a lot of other convenient reasons to prefer this stack. Let’s take a look at some of the benefits of Software 2.0 (think: a ConvNet) compared to Software 1.0 (think: a production-level C++ code base). Software 2.0 is:","hasDropCap":null,"dropCapImage":null,"markups":[],"codeBlockMetadata":null,"iframe":null,"mixtapeMetadata":null},"Paragraph:b1853291a8ec_21":{"__typename":"Paragraph","id":"b1853291a8ec_21","name":"4486","type":"P","href":null,"layout":null,"metadata":null,"text":"Computationally homogeneous. A typical neural network is, to the first order, made up of a sandwich of only two operations: matrix multiplication and thresholding at zero (ReLU). Compare that with the instruction set of classical software, which is significantly more heterogenous and complex. Because you only have to provide Software 1.0 implementation for a small number of the core computational primitives (e.g. matrix multiply), it is much easier to make various correctness\u002Fperformance guarantees.","hasDropCap":null,"dropCapImage":null,"markups":[{"__typename":"Markup","type":"STRONG","start":0,"end":27,"href":null,"anchorType":null,"userId":null,"linkMetadata":null}],"codeBlockMetadata":null,"iframe":null,"mixtapeMetadata":null},"Paragraph:b1853291a8ec_22":{"__typename":"Paragraph","id":"b1853291a8ec_22","name":"061b","type":"P","href":null,"layout":null,"metadata":null,"text":"Simple to bake into silicon. As a corollary, since the instruction set of a neural network is relatively small, it is significantly easier to implement these networks much closer to silicon, e.g. with custom ASICs, neuromorphic chips, and so on. The world will change when low-powered intelligence becomes pervasive around us. E.g., small, inexpensive chips could come with a pretrained ConvNet, a speech recognizer, and a WaveNet speech synthesis network all integrated in a small protobrain that you can attach to stuff.","hasDropCap":null,"dropCapImage":null,"markups":[{"__typename":"Markup","type":"A","start":208,"end":213,"href":"https:\u002F\u002Fwww.forbes.com\u002Fsites\u002Fmoorinsights\u002F2017\u002F08\u002F04\u002Fwill-asic-chips-become-the-next-big-thing-in-ai\u002F#7d6d7c0511d9","anchorType":"LINK","userId":null,"linkMetadata":null},{"__typename":"Markup","type":"A","start":215,"end":233,"href":"https:\u002F\u002Fspectrum.ieee.org\u002Fsemiconductors\u002Fdesign\u002Fneuromorphic-chips-are-destined-for-deep-learningor-obscurity","anchorType":"LINK","userId":null,"linkMetadata":null},{"__typename":"Markup","type":"STRONG","start":0,"end":27,"href":null,"anchorType":null,"userId":null,"linkMetadata":null}],"codeBlockMetadata":null,"iframe":null,"mixtapeMetadata":null},"Paragraph:b1853291a8ec_23":{"__typename":"Paragraph","id":"b1853291a8ec_23","name":"b1d9","type":"P","href":null,"layout":null,"metadata":null,"text":"Constant running time. Every iteration of a typical neural net forward pass takes exactly the same amount of FLOPS. There is zero variability based on the different execution paths your code could take through some sprawling C++ code base. Of course, you could have dynamic compute graphs but the execution flow is normally still significantly constrained. This way we are also almost guaranteed to never find ourselves in unintended infinite loops.","hasDropCap":null,"dropCapImage":null,"markups":[{"__typename":"Markup","type":"STRONG","start":0,"end":21,"href":null,"anchorType":null,"userId":null,"linkMetadata":null}],"codeBlockMetadata":null,"iframe":null,"mixtapeMetadata":null},"Paragraph:b1853291a8ec_24":{"__typename":"Paragraph","id":"b1853291a8ec_24","name":"c82b","type":"P","href":null,"layout":null,"metadata":null,"text":"Constant memory use. Related to the above, there is no dynamically allocated memory anywhere so there is also little possibility of swapping to disk, or memory leaks that you have to hunt down in your code.","hasDropCap":null,"dropCapImage":null,"markups":[{"__typename":"Markup","type":"STRONG","start":0,"end":19,"href":null,"anchorType":null,"userId":null,"linkMetadata":null}],"codeBlockMetadata":null,"iframe":null,"mixtapeMetadata":null},"Paragraph:b1853291a8ec_25":{"__typename":"Paragraph","id":"b1853291a8ec_25","name":"3424","type":"P","href":null,"layout":null,"metadata":null,"text":"It is highly portable. A sequence of matrix multiplies is significantly easier to run on arbitrary computational configurations compared to classical binaries or scripts.","hasDropCap":null,"dropCapImage":null,"markups":[{"__typename":"Markup","type":"STRONG","start":0,"end":21,"href":null,"anchorType":null,"userId":null,"linkMetadata":null}],"codeBlockMetadata":null,"iframe":null,"mixtapeMetadata":null},"Paragraph:b1853291a8ec_26":{"__typename":"Paragraph","id":"b1853291a8ec_26","name":"febb","type":"P","href":null,"layout":null,"metadata":null,"text":"It is very agile. If you had a C++ code and someone wanted you to make it twice as fast (at cost of performance if needed), it would be highly non-trivial to tune the system for the new spec. However, in Software 2.0 we can take our network, remove half of the channels, retrain, and there — it runs exactly at twice the speed and works a bit worse. It’s magic. Conversely, if you happen to get more data\u002Fcompute, you can immediately make your program work better just by adding more channels and retraining.","hasDropCap":null,"dropCapImage":null,"markups":[{"__typename":"Markup","type":"STRONG","start":0,"end":16,"href":null,"anchorType":null,"userId":null,"linkMetadata":null}],"codeBlockMetadata":null,"iframe":null,"mixtapeMetadata":null},"Paragraph:b1853291a8ec_27":{"__typename":"Paragraph","id":"b1853291a8ec_27","name":"075d","type":"P","href":null,"layout":null,"metadata":null,"text":"Modules can meld into an optimal whole. Our software is often decomposed into modules that communicate through public functions, APIs, or endpoints. However, if two Software 2.0 modules that were originally trained separately interact, we can easily backpropagate through the whole. Think about how amazing it could be if your web browser could automatically re-design the low-level system instructions 10 stacks down to achieve a higher efficiency in loading web pages. Or if the computer vision library (e.g. OpenCV) you imported could be auto-tuned on your specific data. With 2.0, this is the default behavior.","hasDropCap":null,"dropCapImage":null,"markups":[{"__typename":"Markup","type":"STRONG","start":0,"end":38,"href":null,"anchorType":null,"userId":null,"linkMetadata":null}],"codeBlockMetadata":null,"iframe":null,"mixtapeMetadata":null},"Paragraph:b1853291a8ec_28":{"__typename":"Paragraph","id":"b1853291a8ec_28","name":"c472","type":"P","href":null,"layout":null,"metadata":null,"text":"It is better than you. Finally, and most importantly, a neural network is a better piece of code than anything you or I can come up with in a large fraction of valuable verticals, which currently at the very least involve anything to do with images\u002Fvideo and sound\u002Fspeech.","hasDropCap":null,"dropCapImage":null,"markups":[{"__typename":"Markup","type":"STRONG","start":0,"end":21,"href":null,"anchorType":null,"userId":null,"linkMetadata":null}],"codeBlockMetadata":null,"iframe":null,"mixtapeMetadata":null},"Paragraph:b1853291a8ec_29":{"__typename":"Paragraph","id":"b1853291a8ec_29","name":"375e","type":"H3","href":null,"layout":null,"metadata":null,"text":"The limitations of Software 2.0","hasDropCap":null,"dropCapImage":null,"markups":[],"codeBlockMetadata":null,"iframe":null,"mixtapeMetadata":null},"Paragraph:b1853291a8ec_30":{"__typename":"Paragraph","id":"b1853291a8ec_30","name":"8f48","type":"P","href":null,"layout":null,"metadata":null,"text":"The 2.0 stack also has some of its own disadvantages. At the end of the optimization we’re left with large networks that work well, but it’s very hard to tell how. Across many applications areas, we’ll be left with a choice of using a 90% accurate model we understand, or 99% accurate model we don’t.","hasDropCap":null,"dropCapImage":null,"markups":[],"codeBlockMetadata":null,"iframe":null,"mixtapeMetadata":null},"Paragraph:b1853291a8ec_31":{"__typename":"Paragraph","id":"b1853291a8ec_31","name":"de95","type":"P","href":null,"layout":null,"metadata":null,"text":"The 2.0 stack can fail in unintuitive and embarrassing ways ,or worse, they can “silently fail”, e.g., by silently adopting biases in their training data, which are very difficult to properly analyze and examine when their sizes are easily in the millions in most cases.","hasDropCap":null,"dropCapImage":null,"markups":[{"__typename":"Markup","type":"A","start":18,"end":59,"href":"https:\u002F\u002Fmotherboard.vice.com\u002Fen_us\u002Farticle\u002Fnz7798\u002Fweve-already-taught-artificial-intelligence-to-be-racist-sexist","anchorType":"LINK","userId":null,"linkMetadata":null}],"codeBlockMetadata":null,"iframe":null,"mixtapeMetadata":null},"Paragraph:b1853291a8ec_32":{"__typename":"Paragraph","id":"b1853291a8ec_32","name":"7188","type":"P","href":null,"layout":null,"metadata":null,"text":"Finally, we’re still discovering some of the peculiar properties of this stack. For instance, the existence of adversarial examples and attacks highlights the unintuitive nature of this stack.","hasDropCap":null,"dropCapImage":null,"markups":[{"__typename":"Markup","type":"A","start":111,"end":131,"href":"https:\u002F\u002Fblog.openai.com\u002Fadversarial-example-research\u002F","anchorType":"LINK","userId":null,"linkMetadata":null},{"__typename":"Markup","type":"A","start":136,"end":143,"href":"https:\u002F\u002Fgithub.com\u002Fyenchenlin\u002Fawesome-adversarial-machine-learning","anchorType":"LINK","userId":null,"linkMetadata":null}],"codeBlockMetadata":null,"iframe":null,"mixtapeMetadata":null},"Paragraph:b1853291a8ec_33":{"__typename":"Paragraph","id":"b1853291a8ec_33","name":"385e","type":"H3","href":null,"layout":null,"metadata":null,"text":"Programming in the 2.0 stack","hasDropCap":null,"dropCapImage":null,"markups":[],"codeBlockMetadata":null,"iframe":null,"mixtapeMetadata":null},"Paragraph:b1853291a8ec_34":{"__typename":"Paragraph","id":"b1853291a8ec_34","name":"3494","type":"P","href":null,"layout":null,"metadata":null,"text":"Software 1.0 is code we write. Software 2.0 is code written by the optimization based on an evaluation criterion (such as “classify this training data correctly”). It is likely that any setting where the program is not obvious but one can repeatedly evaluate the performance of it (e.g. — did you classify some images correctly? do you win games of Go?) will be subject to this transition, because the optimization can find much better code than what a human can write.","hasDropCap":null,"dropCapImage":null,"markups":[],"codeBlockMetadata":null,"iframe":null,"mixtapeMetadata":null},"Paragraph:b1853291a8ec_35":{"__typename":"Paragraph","id":"b1853291a8ec_35","name":"71aa","type":"IMG","href":null,"layout":"INSET_CENTER","metadata":{"__typename":"ImageMetadata","id":"1*7aTCueMW8oBRiqkyobunVA.png","originalHeight":721,"originalWidth":596,"focusPercentX":null,"focusPercentY":null,"alt":null},"text":"","hasDropCap":null,"dropCapImage":null,"markups":[],"codeBlockMetadata":null,"iframe":null,"mixtapeMetadata":null},"Paragraph:b1853291a8ec_36":{"__typename":"Paragraph","id":"b1853291a8ec_36","name":"8133","type":"P","href":null,"layout":null,"metadata":null,"text":"The lens through which we view trends matters. If you recognize Software 2.0 as a new and emerging programming paradigm instead of simply treating neural networks as a pretty good classifier in the class of machine learning techniques, the extrapolations become more obvious, and it’s clear that there is much more work to do.","hasDropCap":null,"dropCapImage":null,"markups":[],"codeBlockMetadata":null,"iframe":null,"mixtapeMetadata":null},"Paragraph:b1853291a8ec_37":{"__typename":"Paragraph","id":"b1853291a8ec_37","name":"df86","type":"P","href":null,"layout":null,"metadata":null,"text":"In particular, we’ve built up a vast amount of tooling that assists humans in writing 1.0 code, such as powerful IDEs with features like syntax highlighting, debuggers, profilers, go to def, git integration, etc. In the 2.0 stack, the programming is done by accumulating, massaging and cleaning datasets. For example, when the network fails in some hard or rare cases, we do not fix those predictions by writing code, but by including more labeled examples of those cases. Who is going to develop the first Software 2.0 IDEs, which help with all of the workflows in accumulating, visualizing, cleaning, labeling, and sourcing datasets? Perhaps the IDE bubbles up images that the network suspects are mislabeled based on the per-example loss, or assists in labeling by seeding labels with predictions, or suggests useful examples to label based on the uncertainty of the network’s predictions.","hasDropCap":null,"dropCapImage":null,"markups":[],"codeBlockMetadata":null,"iframe":null,"mixtapeMetadata":null},"Paragraph:b1853291a8ec_38":{"__typename":"Paragraph","id":"b1853291a8ec_38","name":"c0b6","type":"P","href":null,"layout":null,"metadata":null,"text":"Similarly, Github is a very successful home for Software 1.0 code. Is there space for a Software 2.0 Github? In this case repositories are datasets and commits are made up of additions and edits of the labels.","hasDropCap":null,"dropCapImage":null,"markups":[],"codeBlockMetadata":null,"iframe":null,"mixtapeMetadata":null},"Paragraph:b1853291a8ec_39":{"__typename":"Paragraph","id":"b1853291a8ec_39","name":"5414","type":"P","href":null,"layout":null,"metadata":null,"text":"Traditional package managers and related serving infrastructure like pip, conda, docker, etc. help us more easily deploy and compose binaries. How do we effectively deploy, share, import and work with Software 2.0 binaries? What is the conda equivalent for neural networks?","hasDropCap":null,"dropCapImage":null,"markups":[],"codeBlockMetadata":null,"iframe":null,"mixtapeMetadata":null},"Paragraph:b1853291a8ec_40":{"__typename":"Paragraph","id":"b1853291a8ec_40","name":"9cc6","type":"P","href":null,"layout":null,"metadata":null,"text":"In the short term, Software 2.0 will become increasingly prevalent in any domain where repeated evaluation is possible and cheap, and where the algorithm itself is difficult to design explicitly. There are many exciting opportunities to consider the entire software development ecosystem and how it can be adapted to this new programming paradigm. And in the long run, the future of this paradigm is bright because it is increasingly clear that when we develop AGI, it will certainly be written in Software 2.0.","hasDropCap":null,"dropCapImage":null,"markups":[],"codeBlockMetadata":null,"iframe":null,"mixtapeMetadata":null},"UserViewerEdge:userId:ac9d9a35533e-viewerId:lo_93d1296512d8":{"__typename":"UserViewerEdge","id":"userId:ac9d9a35533e-viewerId:lo_93d1296512d8","isMuting":false},"TagViewerEdge:tagSlug:machine-learning-viewerId:lo_93d1296512d8":{"__typename":"TagViewerEdge","id":"tagSlug:machine-learning-viewerId:lo_93d1296512d8","isMuting":false,"isFollowing":false},"Tag:machine-learning":{"__typename":"Tag","viewerEdge":{"__ref":"TagViewerEdge:tagSlug:machine-learning-viewerId:lo_93d1296512d8"},"id":"machine-learning","displayTitle":"Machine Learning","normalizedTagSlug":"machine-learning"},"TagViewerEdge:tagSlug:artificial-intelligence-viewerId:lo_93d1296512d8":{"__typename":"TagViewerEdge","id":"tagSlug:artificial-intelligence-viewerId:lo_93d1296512d8","isMuting":false,"isFollowing":false},"Tag:artificial-intelligence":{"__typename":"Tag","viewerEdge":{"__ref":"TagViewerEdge:tagSlug:artificial-intelligence-viewerId:lo_93d1296512d8"},"id":"artificial-intelligence","displayTitle":"Artificial Intelligence","normalizedTagSlug":"artificial-intelligence"},"TagViewerEdge:tagSlug:programming-viewerId:lo_93d1296512d8":{"__typename":"TagViewerEdge","id":"tagSlug:programming-viewerId:lo_93d1296512d8","isMuting":false,"isFollowing":false},"Tag:programming":{"__typename":"Tag","viewerEdge":{"__ref":"TagViewerEdge:tagSlug:programming-viewerId:lo_93d1296512d8"},"id":"programming","displayTitle":"Programming","normalizedTagSlug":"programming"},"TagViewerEdge:tagSlug:software-development-viewerId:lo_93d1296512d8":{"__typename":"TagViewerEdge","id":"tagSlug:software-development-viewerId:lo_93d1296512d8","isMuting":false,"isFollowing":false},"Tag:software-development":{"__typename":"Tag","viewerEdge":{"__ref":"TagViewerEdge:tagSlug:software-development-viewerId:lo_93d1296512d8"},"id":"software-development","displayTitle":"Software Development","normalizedTagSlug":"software-development"},"TagViewerEdge:tagSlug:future-viewerId:lo_93d1296512d8":{"__typename":"TagViewerEdge","id":"tagSlug:future-viewerId:lo_93d1296512d8","isMuting":false,"isFollowing":false},"Tag:future":{"__typename":"Tag","viewerEdge":{"__ref":"TagViewerEdge:tagSlug:future-viewerId:lo_93d1296512d8"},"id":"future","displayTitle":"Future","normalizedTagSlug":"future"},"PostViewerEdge:postId:a64152b37c35-viewerId:lo_93d1296512d8":{"__typename":"PostViewerEdge","hasReposted":false,"id":"postId:a64152b37c35-viewerId:lo_93d1296512d8","repostNote":null,"shouldIndexPostForExternalSearch":true},"Post:a64152b37c35":{"__typename":"Post","id":"a64152b37c35","isPublished":true,"allowResponses":true,"isLimitedState":false,"collection":null,"content({\"postMeteringOptions\":{\"referrer\":\"\"}})":{"__typename":"PostContent","isLockedPreviewOnly":false,"bodyModel":{"__typename":"RichText","sections":[{"__typename":"Section","name":"ae17","startIndex":0,"textLayout":null,"imageLayout":null,"backgroundImage":null,"videoLayout":null,"backgroundVideo":null}],"paragraphs":[{"__ref":"Paragraph:b1853291a8ec_0"},{"__ref":"Paragraph:b1853291a8ec_1"},{"__ref":"Paragraph:b1853291a8ec_2"},{"__ref":"Paragraph:b1853291a8ec_3"},{"__ref":"Paragraph:b1853291a8ec_4"},{"__ref":"Paragraph:b1853291a8ec_5"},{"__ref":"Paragraph:b1853291a8ec_6"},{"__ref":"Paragraph:b1853291a8ec_7"},{"__ref":"Paragraph:b1853291a8ec_8"},{"__ref":"Paragraph:b1853291a8ec_9"},{"__ref":"Paragraph:b1853291a8ec_10"},{"__ref":"Paragraph:b1853291a8ec_11"},{"__ref":"Paragraph:b1853291a8ec_12"},{"__ref":"Paragraph:b1853291a8ec_13"},{"__ref":"Paragraph:b1853291a8ec_14"},{"__ref":"Paragraph:b1853291a8ec_15"},{"__ref":"Paragraph:b1853291a8ec_16"},{"__ref":"Paragraph:b1853291a8ec_17"},{"__ref":"Paragraph:b1853291a8ec_18"},{"__ref":"Paragraph:b1853291a8ec_19"},{"__ref":"Paragraph:b1853291a8ec_20"},{"__ref":"Paragraph:b1853291a8ec_21"},{"__ref":"Paragraph:b1853291a8ec_22"},{"__ref":"Paragraph:b1853291a8ec_23"},{"__ref":"Paragraph:b1853291a8ec_24"},{"__ref":"Paragraph:b1853291a8ec_25"},{"__ref":"Paragraph:b1853291a8ec_26"},{"__ref":"Paragraph:b1853291a8ec_27"},{"__ref":"Paragraph:b1853291a8ec_28"},{"__ref":"Paragraph:b1853291a8ec_29"},{"__ref":"Paragraph:b1853291a8ec_30"},{"__ref":"Paragraph:b1853291a8ec_31"},{"__ref":"Paragraph:b1853291a8ec_32"},{"__ref":"Paragraph:b1853291a8ec_33"},{"__ref":"Paragraph:b1853291a8ec_34"},{"__ref":"Paragraph:b1853291a8ec_35"},{"__ref":"Paragraph:b1853291a8ec_36"},{"__ref":"Paragraph:b1853291a8ec_37"},{"__ref":"Paragraph:b1853291a8ec_38"},{"__ref":"Paragraph:b1853291a8ec_39"},{"__ref":"Paragraph:b1853291a8ec_40"}]},"validatedShareKey":"","shareKeyCreator":null},"creator":{"__ref":"User:ac9d9a35533e"},"inResponseToEntityType":null,"isLocked":false,"isMarkedPaywallOnly":false,"lockedSource":"LOCKED_POST_SOURCE_NONE","mediumUrl":"https:\u002F\u002Fkarpathy.medium.com\u002Fsoftware-2-0-a64152b37c35","primaryTopic":{"__ref":"Topic:decb52b64abf"},"topics":[{"__typename":"Topic","slug":"artificial-intelligence"},{"__typename":"Topic","slug":"machine-learning"},{"__typename":"Topic","slug":"software-engineering"},{"__typename":"Topic","slug":"programming"}],"responsesLocked":false,"visibility":"PUBLIC","latestPublishedVersion":"b1853291a8ec","postResponses":{"__typename":"PostResponses","count":192},"responseDistribution":"NOT_DISTRIBUTED","clapCount":61014,"title":"Software 2.0","isSeries":false,"sequence":null,"uniqueSlug":"software-2-0-a64152b37c35","tags":[{"__ref":"Tag:machine-learning"},{"__ref":"Tag:artificial-intelligence"},{"__ref":"Tag:programming"},{"__ref":"Tag:software-development"},{"__ref":"Tag:future"}],"inResponseToPostResult":null,"inResponseToCatalogResult":null,"viewerEdge":{"__ref":"PostViewerEdge:postId:a64152b37c35-viewerId:lo_93d1296512d8"},"extendedPreviewContent":{"__typename":"PreviewContent","subtitle":"I sometimes see people refer to neural networks as just “another tool in your machine learning toolbox”. They have some pros and cons, they…","isFullContent":false},"previewImage":{"__typename":"ImageMetadata","id":"1*CHcu2L0NmAZwCpQgmS1ByA.jpeg","focusPercentX":null,"focusPercentY":null,"alt":null,"originalHeight":234,"originalWidth":1017},"firstPublishedAt":1510438733751,"latestPublishedAt":1615659523264,"pinnedAt":0,"socialTitle":"","socialDek":"","canonicalUrl":"https:\u002F\u002Fkarpathy.medium.com\u002Fsoftware-2-0-a64152b37c35","metaDescription":"","readingTime":8.79811320754717,"previewContent":{"__typename":"PreviewContent","subtitle":"I sometimes see people refer to neural networks as just “another tool in your machine learning toolbox”. They have some pros and cons, they…"},"isShortform":false,"seoMetaTags":{"__typename":"SEOMetaTags","robots":"index,follow","jsonLd":"{\"@context\":\"https:\u002F\u002Fschema.org\",\"@id\":\"https:\u002F\u002Fkarpathy.medium.com\u002Fsoftware-2-0-a64152b37c35\",\"@type\":\"SocialMediaPosting\",\"image\":[\"https:\u002F\u002Fmiro.medium.com\u002F1*CHcu2L0NmAZwCpQgmS1ByA.jpeg\"],\"url\":\"https:\u002F\u002Fkarpathy.medium.com\u002Fsoftware-2-0-a64152b37c35\",\"dateCreated\":\"2017-11-11T22:18:53Z\",\"datePublished\":\"2017-11-11T22:18:53Z\",\"dateModified\":\"2021-03-13T18:18:43Z\",\"headline\":\"Software 2.0\",\"name\":\"Software 2.0\",\"description\":\"Software 2.0 I sometimes see people refer to neural networks as just “another tool in your machine learning toolbox”. They have some pros and cons, they work here or there, and sometimes you can …\",\"identifier\":\"a64152b37c35\",\"author\":{\"@context\":\"https:\u002F\u002Fschema.org\",\"@id\":\"https:\u002F\u002Fmedium.com\u002F@karpathy\",\"@type\":\"Person\",\"identifier\":\"karpathy\",\"name\":\"Andrej Karpathy\",\"url\":\"https:\u002F\u002Fmedium.com\u002F@karpathy\"},\"creator\":{\"@context\":\"https:\u002F\u002Fschema.org\",\"@id\":\"https:\u002F\u002Fmedium.com\u002F@karpathy\",\"@type\":\"Person\",\"identifier\":\"karpathy\",\"name\":\"Andrej Karpathy\",\"url\":\"https:\u002F\u002Fmedium.com\u002F@karpathy\"},\"publisher\":{\"@context\":\"https:\u002F\u002Fschema.org\",\"@type\":\"Organization\",\"@id\":\"https:\u002F\u002Fmedium.com\",\"name\":\"Medium\",\"url\":\"https:\u002F\u002Fmedium.com\",\"logo\":{\"@type\":\"ImageObject\",\"width\":500,\"height\":110,\"url\":\"https:\u002F\u002Fmiro.medium.com\u002Fv2\u002Fresize:fit:500\u002F7%2AV1_7XP4snlmqrc_0Njontw.png\"}},\"mainEntityOfPage\":\"https:\u002F\u002Fkarpathy.medium.com\u002Fsoftware-2-0-a64152b37c35\",\"isAccessibleForFree\":true}"},"seoDescription":"Software 2.0 I sometimes see people refer to neural networks as just “another tool in your machine learning toolbox”. They have some pros and cons, they work here or there, and sometimes you can …","shortformType":"SHORTFORM_TYPE_LINK","seoTitle":"","isSuspended":false,"license":"ALL_RIGHTS_RESERVED","isFeaturedInPublishedPublication":false,"isNewsletter":false,"statusForCollection":null,"pendingCollection":null,"detectedLanguage":"en","wordCount":2146}};</script><script>window.__GRAPHQL_URI__ = "\u002F_\u002Fgraphql";</script><script>window.__PRELOADED_STATE__ = {"algolia":{"queries":{}},"cache":{"experimentGroupSet":true,"reason":"","group":"enabled","tags":["group-edgeCachePosts","group-enableDarkModeV0","group-enableTagPillsAtStoryTop","post-a64152b37c35","user-ac9d9a35533e"],"serverVariantState":"44136fa355b3678a1146ad16f7e8649e94fb4fc21fe77e8310c060f61caaff8a","middlewareEnabled":true,"cacheStatus":"DYNAMIC","shouldUseCache":true,"vary":[],"enableDarkModeV0":false,"tagPillsWithPopoverEnabled":"control","tagPillsAtStoryTopEnabled":"control","membershipOffersEnabled":false},"client":{"hydrated":false,"isUs":false,"isNativeMedium":false,"isSafariMobile":false,"isSafari":false,"isFirefox":false,"routingEntity":{"type":"USER","id":"ac9d9a35533e","explicit":true},"viewerIsBot":false,"clientScore":45},"debug":{"requestId":"00b69dcb-3dd4-4411-9215-10cbef2cce29","requestTag":"","hybridDevServices":[]},"multiVote":{"clapsPerPost":{}},"navigation":{"hideGoogleOneTap":false,"hasRenderedAlternateUserBanner":null,"currentLocation":"https:\u002F\u002Fkarpathy.medium.com\u002Fsoftware-2-0-a64152b37c35","host":"karpathy.medium.com","hostname":"karpathy.medium.com","referrer":"","hasSetReferrer":false,"susiModal":{"step":null,"operation":"register","rememberMe":true},"postRead":false,"paymentFunnel":{},"partnerProgram":{"selectedCountryCode":null},"staticRouterContext":{"route":{"name":"ShowPostUnderUser"},"statusCode":200},"toastQueue":[],"currentToast":null,"queryString":"","currentHash":""},"config":{"nodeEnv":"production","version":"main-20260717-185602-2cae14dcd9","target":"production","productName":"Medium","publicUrl":"https:\u002F\u002Fcdn-client.medium.com\u002Flite","authDomain":"medium.com","authGoogleClientId":"216296035834-k1k6qe060s2tp2a2jam4ljdcms00sttg.apps.googleusercontent.com","favicon":"production","iosAppId":"828256236","glyphUrl":"https:\u002F\u002Fglyph.medium.com","branchKey":"key_live_ofxXr2qTrrU9NqURK8ZwEhknBxiI6KBm","algolia":{"appId":"MQ57UUUQZ2","apiKeySearch":"394474ced050e3911ae2249ecc774921","indexPrefix":"medium_","host":"-dsn.algolia.net"},"recaptchaKey":"6Lfc37IUAAAAAKGGtC6rLS13R1Hrw_BqADfS1LRk","recaptcha3Key":"6Lf8R9wUAAAAABMI_85Wb8melS7Zj6ziuf99Yot5","recaptchaEnterpriseKeyId":"6Le-uGgpAAAAAPprRaokM8AKthQ9KNGdoxaGUvVp","datadog":{"applicationId":"6702d87d-a7e0-42fe-bbcb-95b469547ea0","clientToken":"pub853ea8d17ad6821d9f8f11861d23dfed","rumToken":"pubf9cc52896502b9413b68ba36fc0c7162","context":{"deployment":{"target":"production","tag":"main-20260717-185602-2cae14dcd9","commit":"2cae14dcd90f563afbc593f46312c3b3deff2752"}},"datacenter":"us"},"googleAdsCode":"AW-17106321204","googleAnalyticsCode":"G-7JY7T788PK","googlePay":{"apiVersion":"2","apiVersionMinor":"0","merchantId":"BCR2DN6TV7EMTGBM","merchantName":"Medium","instanceMerchantId":"13685562959212738550"},"applePay":{"version":3},"signInWallCustomDomainCollectionIds":["3a8144eabfe3","336d898217ee","61061eb0c96b","138adf9c44c","819cc2aaeee0"],"mediumMastodonDomainName":"me.dm","mediumOwnedAndOperatedCollectionIds":["8a9336e5bb4","b7e45b22fec3","193b68bd4fba","8d6b8a439e32","54c98c43354d","3f6ecf56618","d944778ce714","92d2092dc598","ae2a65f35510","1285ba81cada","544c7006046e","fc8964313712","40187e704f1c","88d9857e584e","7b6769f2748b","bcc38c8f6edf","cef6983b292","cb8577c9149e","444d13b52878","713d7dbc99b0","ef8e90590e66","191186aaafa0","55760f21cdc5","9dc80918cc93","bdc4052bbdba","8ccfed20cbb2"],"tierOneDomains":["medium.com","thebolditalic.com","arcdigital.media","towardsdatascience.com","uxdesign.cc","codeburst.io","psiloveyou.xyz","writingcooperative.com","entrepreneurshandbook.co","prototypr.io","betterhumans.coach.me","theascent.pub"],"topicsToFollow":["d61cf867d93f","8a146bc21b28","1eca0103fff3","4d562ee63426","aef1078a3ef5","e15e46793f8d","6158eb913466","55f1c20aba7a","3d18b94f6858","4861fee224fd","63c6f1f93ee","1d98b3a9a871","decb52b64abf","ae5d4995e225","830cded25262"],"topicToTagMappings":{"accessibility":"accessibility","addiction":"addiction","android-development":"android-development","art":"art","artificial-intelligence":"artificial-intelligence","astrology":"astrology","basic-income":"basic-income","beauty":"beauty","biotech":"biotech","blockchain":"blockchain","books":"books","business":"business","cannabis":"cannabis","cities":"cities","climate-change":"climate-change","comics":"comics","coronavirus":"coronavirus","creativity":"creativity","cryptocurrency":"cryptocurrency","culture":"culture","cybersecurity":"cybersecurity","data-science":"data-science","design":"design","digital-life":"digital-life","disability":"disability","economy":"economy","education":"education","equality":"equality","family":"family","feminism":"feminism","fiction":"fiction","film":"film","fitness":"fitness","food":"food","freelancing":"freelancing","future":"future","gadgets":"gadgets","gaming":"gaming","gun-control":"gun-control","health":"health","history":"history","humor":"humor","immigration":"immigration","ios-development":"ios-development","javascript":"javascript","justice":"justice","language":"language","leadership":"leadership","lgbtqia":"lgbtqia","lifestyle":"lifestyle","machine-learning":"machine-learning","makers":"makers","marketing":"marketing","math":"math","media":"media","mental-health":"mental-health","mindfulness":"mindfulness","money":"money","music":"music","neuroscience":"neuroscience","nonfiction":"nonfiction","outdoors":"outdoors","parenting":"parenting","pets":"pets","philosophy":"philosophy","photography":"photography","podcasts":"podcast","poetry":"poetry","politics":"politics","privacy":"privacy","product-management":"product-management","productivity":"productivity","programming":"programming","psychedelics":"psychedelics","psychology":"psychology","race":"race","relationships":"relationships","religion":"religion","remote-work":"remote-work","san-francisco":"san-francisco","science":"science","self":"self","self-driving-cars":"self-driving-cars","sexuality":"sexuality","social-media":"social-media","society":"society","software-engineering":"software-engineering","space":"space","spirituality":"spirituality","sports":"sports","startups":"startup","style":"style","technology":"technology","transportation":"transportation","travel":"travel","true-crime":"true-crime","tv":"tv","ux":"ux","venture-capital":"venture-capital","visual-design":"visual-design","work":"work","world":"world","writing":"writing"},"defaultImages":{"avatar":{"imageId":"1*dmbNkD5D-u45r44go_cf0g.png","height":150,"width":150},"orgLogo":{"imageId":"7*V1_7XP4snlmqrc_0Njontw.png","height":110,"width":500},"postLogo":{"imageId":"167cff2a3d17ac1e64d0762539978f2d54c0058886e8b3c8a03a725a83012ec0","height":630,"width":1200},"postPreviewImage":{"imageId":"bc1f8416df0cad099e43cda2872716e5864f18a73bda2a7547ea082aca9b5632","height":630,"width":1200}},"embeddedPostIds":{"coronavirus":"cd3010f9d81f"},"sharedCdcMessaging":{"COVID_APPLICABLE_TAG_SLUGS":[],"COVID_APPLICABLE_TOPIC_NAMES":[],"COVID_APPLICABLE_TOPIC_NAMES_FOR_TOPIC_PAGE":[],"COVID_MESSAGES":{"tierA":{"text":"For more information on the novel coronavirus and Covid-19, visit cdc.gov.","markups":[{"start":66,"end":73,"href":"https:\u002F\u002Fwww.cdc.gov\u002Fcoronavirus\u002F2019-nCoV"}]},"tierB":{"text":"Anyone can publish on Medium per our Policies, but we don’t fact-check every story. For more info about the coronavirus, see cdc.gov.","markups":[{"start":37,"end":45,"href":"https:\u002F\u002Fhelp.medium.com\u002Fhc\u002Fen-us\u002Fcategories\u002F201931128-Policies-Safety"},{"start":125,"end":132,"href":"https:\u002F\u002Fwww.cdc.gov\u002Fcoronavirus\u002F2019-nCoV"}]},"paywall":{"text":"This article has been made free for everyone, thanks to Medium Members. For more information on the novel coronavirus and Covid-19, visit cdc.gov.","markups":[{"start":56,"end":70,"href":"https:\u002F\u002Fmedium.com\u002Fmembership"},{"start":138,"end":145,"href":"https:\u002F\u002Fwww.cdc.gov\u002Fcoronavirus\u002F2019-nCoV"}]},"unbound":{"text":"This article is free for everyone, thanks to Medium Members. For more information on the novel coronavirus and Covid-19, visit cdc.gov.","markups":[{"start":45,"end":59,"href":"https:\u002F\u002Fmedium.com\u002Fmembership"},{"start":127,"end":134,"href":"https:\u002F\u002Fwww.cdc.gov\u002Fcoronavirus\u002F2019-nCoV"}]}},"COVID_BANNER_POST_ID_OVERRIDE_WHITELIST":["3b31a67bff4a"]},"sharedVoteMessaging":{"TAGS":["politics","election-2020","government","us-politics","election","2020-presidential-race","trump","donald-trump","democrats","republicans","congress","republican-party","democratic-party","biden","joe-biden","maga"],"TOPICS":["politics","election"],"MESSAGE":{"text":"Find out more about the U.S. election results here.","markups":[{"start":46,"end":50,"href":"https:\u002F\u002Fcookpolitical.com\u002F2020-national-popular-vote-tracker"}]},"EXCLUDE_POSTS":["397ef29e3ca5"]},"embedPostRules":[],"recircOptions":{"v1":{"limit":3},"v2":{"limit":8}},"braintreeClientKey":"production_zjkj96jm_m56f8fqpf7ngnrd4","braintree":{"enabled":true,"merchantId":"m56f8fqpf7ngnrd4","merchantAccountId":{"usd":"AMediumCorporation_instant","eur":"amediumcorporation_EUR","cad":"amediumcorporation_CAD"},"publicKey":"ds2nn34bg2z7j5gd","braintreeEnvironment":"production","dashboardUrl":"https:\u002F\u002Fwww.braintreegateway.com\u002Fmerchants","gracePeriodDurationInDays":14,"mediumMembershipPlanId":{"monthly":"ce105f8c57a3","monthlyV2":"e8a5e126-792b-4ee6-8fba-d574c1b02fc5","monthlyWithTrial":"d5ee3dbe3db8","monthlyPremium":"fa741a9b47a2","yearly":"a40ad4a43185","yearlyV2":"3815d7d6-b8ca-4224-9b8c-182f9047866e","yearlyStaff":"d74fb811198a","yearlyWithTrial":"b3bc7350e5c7","yearlyPremium":"e21bd2c12166","monthlyOneYearFree":"e6c0637a-2bad-4171-ab4f-3c268633d83c","monthly25PercentOffFirstYear":"235ecc62-0cdb-49ae-9378-726cd21c504b","monthly20PercentOffFirstYear":"ba518864-9c13-4a99-91ca-411bf0cac756","monthly15PercentOffFirstYear":"594c029b-9f89-43d5-88f8-8173af4e070e","monthly10PercentOffFirstYear":"c6c7bc9a-40f2-4b51-8126-e28511d5bdb0","monthlyForStudents":"629ebe51-da7d-41fd-8293-34cd2f2030a8","yearlyOneYearFree":"78ba7be9-0d9f-4ece-aa3e-b54b826f2bf1","yearly25PercentOffFirstYear":"2dbb010d-bb8f-4eeb-ad5c-a08509f42d34","yearly20PercentOffFirstYear":"47565488-435b-47f8-bf93-40d5fbe0ebc8","yearly15PercentOffFirstYear":"8259809b-0881-47d9-acf7-6c001c7f720f","yearly10PercentOffFirstYear":"9dd694fb-96e1-472c-8d9e-3c868d5c1506","yearlyForStudents":"e29345ef-ab1c-4234-95c5-70e50fe6bc23","monthlyCad":"p52orjkaceei","yearlyCad":"h4q9g2up9ktt"},"braintreeDiscountId":{"oneMonthFree":"MONTHS_FREE_01","threeMonthsFree":"MONTHS_FREE_03","sixMonthsFree":"MONTHS_FREE_06","fiftyPercentOffOneYear":"FIFTY_PERCENT_OFF_ONE_YEAR"},"3DSecureVersion":"2","defaultCurrency":"usd","providerPlanIdCurrency":{"4ycw":"usd","rz3b":"usd","3kqm":"usd","jzw6":"usd","c2q2":"usd","nnsw":"usd","q8qw":"usd","d9y6":"usd","fx7w":"cad","nwf2":"cad"}},"paypalClientId":"AXj1G4fotC2GE8KzWX9mSxCH1wmPE3nJglf4Z2ig_amnhvlMVX87otaq58niAg9iuLktVNF_1WCMnN7v","paypal":{"host":"https:\u002F\u002Fapi.paypal.com:443","clientMode":"production","serverMode":"live","webhookId":"4G466076A0294510S","monthlyPlan":{"planId":"P-9WR0658853113943TMU5FDQA","name":"Medium Membership (Monthly) with setup fee","description":"Unlimited access to the best and brightest stories on Medium. Membership billed monthly."},"yearlyPlan":{"planId":"P-7N8963881P8875835MU5JOPQ","name":"Medium Membership (Annual) with setup fee","description":"Unlimited access to the best and brightest stories on Medium. Membership billed annually."},"oneYearGift":{"name":"Medium Membership (1 Year, Digital Gift Code)","description":"Unlimited access to the best and brightest stories on Medium. Gift codes can be redeemed at medium.com\u002Fredeem.","price":"50.00","currency":"USD","sku":"membership-gift-1-yr"},"oldMonthlyPlan":{"planId":"P-96U02458LM656772MJZUVH2Y","name":"Medium Membership (Monthly)","description":"Unlimited access to the best and brightest stories on Medium. Membership billed monthly."},"oldYearlyPlan":{"planId":"P-59P80963JF186412JJZU3SMI","name":"Medium Membership (Annual)","description":"Unlimited access to the best and brightest stories on Medium. Membership billed annually."},"monthlyPlanWithTrial":{"planId":"P-66C21969LR178604GJPVKUKY","name":"Medium Membership (Monthly) with setup fee","description":"Unlimited access to the best and brightest stories on Medium. Membership billed monthly."},"yearlyPlanWithTrial":{"planId":"P-6XW32684EX226940VKCT2MFA","name":"Medium Membership (Annual) with setup fee","description":"Unlimited access to the best and brightest stories on Medium. Membership billed annually."},"oldMonthlyPlanNoSetupFee":{"planId":"P-4N046520HR188054PCJC7LJI","name":"Medium Membership (Monthly)","description":"Unlimited access to the best and brightest stories on Medium. Membership billed monthly."},"oldYearlyPlanNoSetupFee":{"planId":"P-7A4913502Y5181304CJEJMXQ","name":"Medium Membership (Annual)","description":"Unlimited access to the best and brightest stories on Medium. Membership billed annually."},"sdkUrl":"https:\u002F\u002Fwww.paypal.com\u002Fsdk\u002Fjs"},"stripePublishableKey":"pk_live_7FReX44VnNIInZwrIIx6ghjl","log":{"json":true,"level":"info"},"imageUploadMaxSizeMb":25,"staffPicks":{"title":"Staff Picks","catalogId":"c7bc6e1ee00f"}},"session":{"xsrf":""}};</script><script>window.__BUILD_ID__ = "main-20260717-185602-2cae14dcd9";</script><script>window.__MIDDLEWARE_STATE__={"session":{"xsrf":""},"cache":{"cacheStatus":"UPDATING"}}</script><script>(function(){function c(){var b=a.contentDocument||a.contentWindow.document;if(b){var d=b.createElement('script');d.innerHTML="window.__CF$cv$params={r:'a1eb069b9cfc35ae',t:'MTc4NDY0NTUwOA=='};var a=document.createElement('script');a.src='/cdn-cgi/challenge-platform/scripts/jsd/main.js';document.getElementsByTagName('head')[0].appendChild(a);";b.getElementsByTagName('head')[0].appendChild(d)}}if(document.body){var a=document.createElement('iframe');a.height=1;a.width=1;a.style.position='absolute';a.style.top=0;a.style.left=0;a.style.border='none';a.style.visibility='hidden';document.body.appendChild(a);if('loading'!==document.readyState)c();else if(window.addEventListener)document.addEventListener('DOMContentLoaded',c);else{var e=document.onreadystatechange||function(){};document.onreadystatechange=function(b){e(b);'loading'!==document.readyState&&(document.onreadystatechange=e,c())}}}})();</script></body></html>Related skills
| Skill | Stars | Updated | Works with | Status | Install |
|---|---|---|---|---|---|
| slack-gif-creatoranthropics/skillsKnowledge and utilities for creating animated GIFs optimized for Slack. Provides constraints, validation tools, and animation concepts. Use when users request animated GIFs for Slack like "make me a GIF of X doing Y for Slack." | Stars: 198k | Updated | Claude CodeCodex | Verified | Install |
| ponytailDietrichGebert/ponytailLazy senior dev mode for any coding task: the smallest change that fully solves it, and a reply a busy human understands in one read. | Stars: 194k | Updated | Claude CodeCodex | Verified | Install |
| ponytail-debtDietrichGebert/ponytailHarvest every shortcut: comment into one debt ledger, so deferrals get tracked instead of forgotten. One-shot report. | Stars: 194k | Updated | Claude CodeCodex | Verified | Install |
| ponytail-debtDietrichGebert/ponytailList every `shortcut:` comment (and older `ponytail:` ones) as a debt ledger. One-shot report, changes nothing. Use for "ponytail debt", "what did ponytail defer", "list the shortcuts", /ponytail-debt. | Stars: 194k | Updated | Claude CodeCodex | Verified | Install |