donaldfilimon/abi / .agents/skills/pptx
pptx
UnverifiedUse this skill any time a .pptx file is involved in any way — as input, output, or both. This includes: creating slide decks, pitch decks, or presentations; reading, parsing, or extracting text from any .pptx file (even if the extracted content will be used elsewhere, like in an email or summary); editing, modifying, or updating existing presentations; combining or splitting slide files; working with templates, layouts, speaker notes, or comments. Trigger whenever the user mentions "deck," "slides," "presentation," or references a .pptx filename, regardless of what they plan to do with the content afterward. If a .pptx file needs to be opened, created, or touched, use this skill.
references/templates/ai_llm_technical_history.js
View on GitHub/**
* class2-replica.js
*
* Recreates "class2.pptx" — A Short History of LLMs — using pptxgenjs.
*
* The deck is 11 slides, 20" x 11.25" (a custom widescreen layout).
* Design: warm paper background (#EFE9DD), Georgia serif headlines,
* Helvetica Neue body, Courier New for monospaced captions, with a
* gold accent (#C6A46A / #A8864A) and dark brown (#3A2C1A).
*
* Four slides (1, 3, 5, 10) use illustration images that live alongside
* this script in ./assets/. If those files are missing the script will
* still run; those illustrations will simply be skipped.
*
* Run: node class2-replica.js
* Out: ./class2-replica.pptx
*/
const path = require("path");
const fs = require("fs");
const pptxgen = require("pptxgenjs");
// ─────────────────────────────────────────────────────────────────────────────
// Theme tokens (pulled directly from the source .pptx XML)
// ─────────────────────────────────────────────────────────────────────────────
const C = {
bg: "EFE9DD", // paper
ink: "3A2C1A", // near-black brown (headlines)
body: "6B5A44", // muted brown (body copy)
meta: "9A8A72", // light taupe (captions / chapter labels)
gold: "A8864A", // deep gold (stat numbers, accents)
goldHi: "C6A46A", // lighter gold (rule lines)
rule: "C9BDA4", // pale rule line
panel: "E6DFCE", // soft panel fill
panelHi: "FCFAF4", // white-ish chat panel
};
const F = {
serif: "Georgia",
sans: "Helvetica Neue",
mono: "Courier New",
};
// Asset directory (for the 4 illustration PNGs from the source deck)
const ASSETS = path.join(__dirname, "assets");
const hasAsset = (name) => fs.existsSync(path.join(ASSETS, name));
const asset = (name) => path.join(ASSETS, name);
// ─────────────────────────────────────────────────────────────────────────────
// Presentation setup — custom 20" x 11.25" layout
// ─────────────────────────────────────────────────────────────────────────────
const pres = new pptxgen();
pres.title = "A Short History of LLMs";
pres.author = "class2 replica";
pres.defineLayout({ name: "CLASS2", width: 20, height: 11.25 });
pres.layout = "CLASS2";
// ─────────────────────────────────────────────────────────────────────────────
// Helpers
// ─────────────────────────────────────────────────────────────────────────────
/** Convert XML `spc` (hundredths of a point) to pptxgenjs `charSpacing` (points). */
const spc = (v) => (v ? v / 100 : 0);
/** Convert XML `sz` (hundredths of a point) to pptxgenjs `fontSize` (points). */
const sz = (v) => v / 100;
/** Add a thin solid bar (rectangle). */
function bar(slide, x, y, w, h, color) {
slide.addShape(pres.shapes.RECTANGLE, {
x, y, w, h,
fill: { color },
line: { type: "none" },
});
}
/** Add a simple left-aligned single-line text box with zero internal padding. */
function label(slide, text, x, y, w, h, opts = {}) {
slide.addText(text, {
x, y, w, h,
fontFace: opts.face || F.sans,
fontSize: opts.size || 12,
color: opts.color || C.body,
bold: opts.bold || false,
italic: opts.italic || false,
align: opts.align || "left",
valign: opts.valign || "top",
charSpacing: opts.charSpacing || 0,
margin: opts.margin != null ? opts.margin : 0.03,
});
}
/** Chapter label (top left) + page counter (top right) + gold rule under it. */
function addHeader(slide, chapter, counter) {
label(slide, chapter, 1.25, 0.5, 6.5, 0.3056, {
face: F.sans, size: 16.5, color: C.meta, charSpacing: 2.97,
});
label(slide, counter, 17.7604, 0.5, 1.0729, 0.3056, {
face: F.sans, size: 16.5, color: C.meta, charSpacing: 2.97,
});
// Gold rule that sits above the slide's headline
bar(slide, 1.25, 1.875, 1.25, 0.0312, C.goldHi);
}
/** Headline (the big Georgia serif sentence).
* Source XML: 54 pt with normAutofit. We drop to 46 pt to fit the 0.85" box. */
function addHeadline(slide, text) {
slide.addText(text, {
x: 1.25, y: 2.2396, w: 18.025, h: 0.8516,
fontFace: F.serif, fontSize: 46, color: C.ink,
charSpacing: -0.81, margin: 0.03,
paraSpaceBefore: 0, paraSpaceAfter: 0,
valign: "top",
});
}
/** Plain paragraph of body copy.
* Note: the source XML specifies 25.5 pt with `normAutofit`, which PowerPoint
* renders smaller to fit the box. pptxgenjs has no equivalent, so we emit
* the text at ~20 pt with a tighter line-height so the box height from the
* original file still holds. */
function addBody(slide, text, x, y, w, h, opts = {}) {
slide.addText(text, {
x, y, w, h,
fontFace: F.sans,
fontSize: opts.size || 20,
color: opts.color || C.body,
lineSpacingMultiple: opts.lineSpacing || 1.35,
margin: 0.03,
valign: "top",
});
}
// ─────────────────────────────────────────────────────────────────────────────
// SLIDE 1 — Title: "A short history of large language models."
// ─────────────────────────────────────────────────────────────────────────────
function buildSlide1() {
const s = pres.addSlide();
s.background = { color: C.bg };
// Top label "A SHORT HISTORY OF LLMS"
label(s, "A SHORT HISTORY OF LLMS", 1.25, 0.5, 4.1824, 0.3056, {
face: F.sans, size: 16.5, color: C.meta, charSpacing: 2.97,
});
// "── A TECHNICAL HISTORY ──"
label(s, "── A TECHNICAL HISTORY ──", 1.25, 1.3229, 9.4417, 0.375, {
face: F.mono, size: 21, color: C.meta, charSpacing: 0.84,
});
// Main title with mixed Georgia / italic gold runs.
// Source: 105 pt in a 9.44 × 7.19 box. At ~88 pt it still wraps to four
// lines but stops short of the date footer at y=9.8".
s.addText(
[
{ text: "A short history of ", options: { color: C.ink, italic: false } },
{ text: "large language models.", options: { color: C.gold, italic: true } },
],
{
x: 1.25, y: 2.0729, w: 9.4417, h: 7.1,
fontFace: F.serif, fontSize: 88,
charSpacing: -2, lineSpacingMultiple: 1.0,
margin: 0.03, valign: "top",
}
);
// Gold rule + date tagline
bar(s, 1.25, 9.953, 0.8333, 0.0312, C.goldHi);
label(s, "1948 — 2026", 2.4167, 9.802, 2.137, 0.375, {
face: F.mono, size: 21, color: C.meta, charSpacing: 0.84,
});
// Illustration (concentric-timeline circle)
if (hasAsset("image-1-1.png")) {
s.addImage({ path: asset("image-1-1.png"), x: 11.25, y: 1.6667, w: 7.5, h: 8.125 });
}
}
// ─────────────────────────────────────────────────────────────────────────────
// SLIDE 2 — Statistics: histogram of P(word | "the cat sat on the")
// ─────────────────────────────────────────────────────────────────────────────
function buildSlide2() {
const s = pres.addSlide();
s.background = { color: C.bg };
addHeader(s, "CHAPTER I · STATISTICS", "02 / 11");
addHeadline(s, "Language modeling began with counting words.");
// Left column: two body paragraphs
addBody(s,
"For half a century, a language model was a table of conditional " +
"probabilities — P(word | previous n-1 words) — estimated by counting " +
"occurrences in a corpus.",
1.25, 4.5332, 8.0724, 2.0955
);
addBody(s,
"Shannon proposed the formulation in 1948. It powered autocomplete, " +
"speech recognition, and machine translation until deep learning arrived.",
1.25, 6.9204, 8.0724, 1.582
);
// Three stat pairs at bottom-left: (1948 / Shannon), (n≤5 / practical limit), (O(Vⁿ) / memory)
const stats = [
{ x: 1.25, big: "1948", lbl: "Shannon", bigW: 1.3086, lblW: 1.3086 },
{ x: 2.9753, big: "n≤5", lbl: "practical limit", bigW: 2.7088, lblW: 2.7088 },
{ x: 6.1007, big: "O(Vⁿ)", lbl: "memory", bigW: 1.3335, lblW: 1.1336 },
];
for (const st of stats) {
label(s, st.big, st.x, 8.9607, st.bigW, 0.5139, {
face: F.mono, size: 30, color: C.gold,
});
label(s, st.lbl, st.x, 9.4329, st.lblW, 0.375, {
face: F.mono, size: 21, color: C.meta,
});
}
// Right panel — the probability histogram
const panelX = 10.1289, panelY = 5.1949, panelW = 8.6211, panelH = 3.9097;
bar(s, panelX, panelY, panelW, panelH, C.panel);
bar(s, panelX, panelY, 0.0417, panelH, C.goldHi); // gold left rule
// Header row: P( ___ | "the cat sat on the" )
label(s, 'P( ___ | "the cat sat on the" )', panelX + 0.5, panelY + 0.4583, 7.8926, 0.3472, {
face: F.mono, size: 17, color: C.ink,
});
// 5 histogram rows: [word, barValue(0-1), prob]
const rows = [
{ word: "mat", val: 0.31, p: "0.31" },
{ word: "floor", val: 0.18, p: "0.18" },
{ word: "chair", val: 0.11, p: "0.11" },
{ word: "roof", val: 0.05, p: "0.05" },
{ word: "…", val: 0.35, p: "0.35" },
];
const rowStartY = 6.2088;
const rowStep = 0.5208;
const barMaxW = 4.0169;
for (let i = 0; i < rows.length; i++) {
const r = rows[i];
const y = rowStartY + i * rowStep;
// word label (left)
label(s, r.word, panelX + 0.5, y, 2.375, 0.3958, {
face: F.mono, size: 19.5, color: C.ink,
});
// bar track (full width)
bar(s, 13.1289, y + 0.0625, barMaxW, 0.2292, C.panel);
// filled bar
bar(s, 13.1289, y + 0.0625, barMaxW * r.val / 0.35 * 0.35 /* keep clean */, 0.2292, C.gold);
// probability value (right)
label(s, r.p, 17.2708, y, 1.0208, 0.3958, {
face: F.mono, size: 19.5, color: C.gold, align: "right",
});
}
}
// Override bar fills for slide 2 with exact widths from XML
function buildSlide2Corrected() { /* kept for reference - using the values above */ }
// ─────────────────────────────────────────────────────────────────────────────
// SLIDE 3 — Geometry: king - man + woman ≈ queen
// ─────────────────────────────────────────────────────────────────────────────
function buildSlide3() {
const s = pres.addSlide();
s.background = { color: C.bg };
addHeader(s, "CHAPTER II · GEOMETRY", "03 / 11");
addHeadline(s, "Word2vec turned meaning into geometry.");
addBody(s,
"In 2013, Mikolov showed that dense vectors, trained to predict " +
"neighboring words, captured analogies as simple arithmetic.",
1.25, 4.3617, 7.725, 1.582
);
// Accent callout box: "king − man + woman ≈ queen"
bar(s, 1.25, 6.3604, 7.5, 1.8665, C.panel);
bar(s, 1.25, 6.3604, 0.0417, 1.8665, C.goldHi);
s.addText(
[
{ text: "king − man + woman ", options: { color: C.ink, italic: true } },
{ text: "≈ queen", options: { color: C.gold, italic: true } },
],
{
x: 1.7083, y: 6.6937, w: 6.85, h: 1.2415,
fontFace: F.serif, fontSize: 42,
valign: "middle", margin: 0.03,
}
);
addBody(s,
"Embeddings — learned, low-dimensional representations — became the " +
"lingua franca of NLP for the decade that followed.",
1.25, 8.6852, 7.725, 1.3102
);
// Right illustration (king / queen / Paris / France / Rome / Italy scatter)
if (hasAsset("image-3-1.png")) {
s.addImage({ path: asset("image-3-1.png"), x: 9.5833, y: 3.6745, w: 9.1667, h: 6.9666 });
}
}
// ─────────────────────────────────────────────────────────────────────────────
// SLIDE 4 — Architecture: 4-stat row + 6x6 self-attention heatmap
// ─────────────────────────────────────────────────────────────────────────────
function buildSlide4() {
const s = pres.addSlide();
s.background = { color: C.bg };
addHeader(s, "CHAPTER III · ARCHITECTURE", "04 / 11");
addHeadline(s, "Attention replaced recurrence.");
addBody(s,
'"Attention Is All You Need" (Vaswani et al., 2017) replaced sequential ' +
"RNNs with a fully parallel architecture built on self-attention.",
1.25, 3.9774, 8.476, 1.582
);
addBody(s,
"The transformer trained faster on more data, scaled almost linearly " +
"with compute, and became the substrate beneath every modern LLM.",
1.25, 5.8511, 8.476, 1.582
);
// Four-stat block (lower-left, 2x2): each item has a rule line, a label, and a note.
const quads = [
{ x: 1.25, y: 7.9748, color: C.rule, label: "RNNS", note: "sequential, hard to parallelize" },
{ x: 5.5312, y: 7.9748, color: C.rule, label: "CNNS", note: "local receptive fields" },
{ x: 1.25, y: 9.1068, color: C.gold, label: "SELF-ATTENTION", note: "global, parallel" },
{ x: 5.5312, y: 9.1068, color: C.gold, label: "CONSEQUENCE", note: "scale unlocked" },
];
for (const q of quads) {
bar(s, q.x, q.y, 3.9479, 0.0208, q.color);
label(s, q.label, q.x, q.y + 0.1667, 4.0664, 0.2778, {
face: F.sans, size: 14.25, color: C.ink, bold: true, charSpacing: 1.5,
});
label(s, q.note, q.x, q.y + 0.4653, 4.0664, 0.375, {
face: F.sans, size: 16.5, color: C.body,
});
}
// ── Right side: self-attention matrix ──
label(s, "SELF-ATTENTION · WEIGHTS", 11.2879, 3.9414, 6.695, 0.2986, {
face: F.sans, size: 14.25, color: C.meta, charSpacing: 2.25, bold: true,
});
const tokens = ["the", "cat", "sat", "on", "the", "mat"];
const headerY = 4.4067;
const rowHs = 0.875;
const colX = [12.6354, 13.5104, 14.3854, 15.2604, 16.1354, 17.0104]; // top-row label x positions
const cellX = [12.6771, 13.5521, 14.4271, 15.3021, 16.1771, 17.0521]; // cell rectangle x positions
const rowStartY = 4.7331;
// Column headers (top row)
for (let c = 0; c < 6; c++) {
label(s, tokens[c], colX[c], headerY, 0.9167, 0.3264, {
face: F.mono, size: 16.5, color: C.meta, align: "center",
});
}
// Row labels + heatmap values (attention weights per row)
const weights = [
[0.60, 0.10, 0.10, 0.05, 0.10, 0.05],
[0.10, 0.50, 0.25, 0.05, 0.05, 0.05],
[0.05, 0.30, 0.40, 0.10, 0.05, 0.10],
[0.05, 0.10, 0.25, 0.40, 0.10, 0.10],
[0.10, 0.05, 0.05, 0.15, 0.50, 0.15],
[0.05, 0.35, 0.10, 0.05, 0.10, 0.35],
];
for (let r = 0; r < 6; r++) {
const y = rowStartY + r * 0.875;
// row label at left
label(s, tokens[r], 11.3021, y + 0.2743, 1.2083, 0.3264, {
face: F.mono, size: 16.5, color: C.meta, align: "right",
});
// cells
for (let c = 0; c < 6; c++) {
const w = weights[r][c];
// cell background (opacity proxies the heatmap — more opaque = larger weight)
s.addShape(pres.shapes.RECTANGLE, {
x: cellX[c], y: y, w: 0.8333, h: 0.8333,
fill: { color: C.gold, transparency: Math.round((1 - w) * 100) },
line: { type: "none" },
});
// cell number centred
label(s, w.toFixed(2), cellX[c] - 0.0347, y + 0.0069, 0.9028, 0.8611, {
face: F.mono, size: 14, color: C.ink, align: "center", valign: "middle",
});
}
}
}
// ─────────────────────────────────────────────────────────────────────────────
// SLIDE 5 — Scale: 1,000× parameter growth
// ─────────────────────────────────────────────────────────────────────────────
function buildSlide5() {
const s = pres.addSlide();
s.background = { color: C.bg };
addHeader(s, "CHAPTER IV · SCALE", "05 / 11");
addHeadline(s, "Scale unlocked emergent abilities.");
addBody(s,
"GPT-3 (2020) scaled the transformer to 175B parameters and demonstrated " +
"in-context learning — solving tasks from a handful of examples, with no " +
"gradient updates.",
1.25, 3.9776, 7.725, 2.0955
);
addBody(s,
"Kaplan's scaling laws made the bet explicit: loss falls as a power law " +
"in parameters, data, and compute.",
1.25, 6.3648, 7.725, 1.582
);
// Tagline: "1,000× parameter growth, 2018 → 2023"
s.addText(
[
{ text: "1,000× ", options: { color: C.ink, italic: true, fontFace: F.serif, fontSize: 48 } },
{ text: "parameter growth, 2018 → 2023", options: { color: C.meta, italic: false, fontFace: F.sans, fontSize: 16 } },
],
{ x: 1.25, y: 8.4052, w: 7.725, h: 1.5417, margin: 0.03, valign: "middle" }
);
// Right illustration: log-scale growth curve
if (hasAsset("image-5-1.png")) {
s.addImage({ path: asset("image-5-1.png"), x: 9.5833, y: 3.7122, w: 9.1667, h: 6.4583 });
}
}
// ─────────────────────────────────────────────────────────────────────────────
// SLIDE 6 — Product: ChatGPT + timeline + mock chat UI
// ─────────────────────────────────────────────────────────────────────────────
function buildSlide6() {
const s = pres.addSlide();
s.background = { color: C.bg };
addHeader(s, "CHAPTER V · PRODUCT", "06 / 11");
addHeadline(s, "ChatGPT made LLMs a consumer product.");
addBody(s,
"InstructGPT (2022) added a second training phase — RLHF — that taught " +
"base models to follow instructions and refuse. ChatGPT wrapped it in a chat UI.",
1.25, 4.4703, 9.3236, 1.582
);
addBody(s,
"The shift from completion engine to assistant was, for most of the " +
"world, the moment LLMs arrived.",
1.25, 6.344, 9.3236, 1.0686
);
// Three-stat launch-timeline row
const items = [
{ x: 1.25, top: "NOV 30", big: "2022", note: "ChatGPT launch" },
{ x: 4.3507, top: "5 DAYS", big: "→ 1M", note: "users" },
{ x: 7.4514, top: "2 MONTHS", big: "→ 100M", note: "users — fastest ever" },
];
for (const it of items) {
bar(s, it.x, 7.9542, 2.8507, 0.0208, C.gold);
label(s, it.top, it.x, 8.1417, 2.9362, 0.2778, {
face: F.mono, size: 15, color: C.meta, charSpacing: 1.5,
});
label(s, it.big, it.x, 8.4612, 2.9362, 0.625, {
face: F.serif, size: 42, color: C.ink,
});
label(s, it.note, it.x, 9.1487, 2.9362, 0.3056, {
face: F.sans, size: 16.5, color: C.body,
});
}
// ── Chat UI panel (right) ──
bar(s, 11.3438, 4.6549, 7.4062, 4.5729, C.panelHi);
// Traffic-light dots
const dots = [
{ x: 11.7674, color: "D06B5B" },
{ x: 11.9965, color: "D9B961" },
{ x: 12.2257, color: "7FA86E" },
];
for (const d of dots) {
s.addShape(pres.shapes.OVAL, {
x: d.x, y: 5.1306, w: 0.1042, h: 0.1042,
fill: { color: d.color }, line: { type: "none" },
});
}
label(s, "CHAT · 2022", 12.6215, 5.0786, 1.5273, 0.25, {
face: F.mono, size: 13.5, color: C.meta, charSpacing: 1.5,
});
// User message line
s.addText(
[
{ text: "User ", options: { bold: true, color: C.ink } },
{ text: "Explain RLHF simply.", options: { color: C.body } },
],
{
x: 11.7674, y: 5.5786, w: 6.7558, h: 0.375,
fontFace: F.sans, fontSize: 21, margin: 0.03, valign: "top",
}
);
// Assistant message
s.addText(
[
{ text: "Assistant ", options: { bold: true, color: C.ink, fontFace: F.sans } },
{ text: "A base model predicts the next token. RLHF teaches it ", options: { color: C.body, fontFace: F.sans } },
{ text: "which ", options: { italic: true, color: C.gold, fontFace: F.serif } },
{ text: "next tokens humans actually prefer — turning a completion engine into an assistant.", options: { color: C.body, fontFace: F.sans } },
],
{
x: 11.7674, y: 6.1202, w: 6.7558, h: 1.7917,
fontSize: 21, lineSpacingMultiple: 1.35, margin: 0.03, valign: "top",
}
);
// Divider + cursor
bar(s, 11.7674, 8.2036, 6.559, 0.0104, C.rule);
label(s, "▍", 11.7674, 8.4223, 6.7558, 0.4236, {
face: F.mono, size: 18, color: C.meta,
});
}
// ─────────────────────────────────────────────────────────────────────────────
// SLIDE 7 — Distribution: open-weights timeline table
// ─────────────────────────────────────────────────────────────────────────────
function buildSlide7() {
const s = pres.addSlide();
s.background = { color: C.bg };
addHeader(s, "CHAPTER VI · DISTRIBUTION", "07 / 11");
addHeadline(s, "Open weights reshaped the field.");
addBody(s,
"After GPT-3, frontier models were API-only. Meta's LLaMA series — " +
"followed by Mistral and DeepSeek — put competitive weights in the hands " +
"of every researcher.",
1.25, 5.0683, 6.4375, 2.6089
);
addBody(s,
"The gap between closed and open frontiers has narrowed from years to months.",
1.25, 7.9689, 6.4375, 0.8874
);
// ── Table (right side) ──
// Header row
const colX = { model: 8.3333, org: 10.8883, rel: 12.6572, sig: 14.426 };
const colW = { model: 2.6383, org: 1.8522, rel: 1.8521, sig: 4.4536 };
const headerY = 4.4227;
// Top border (darker)
bar(s, 8.3333, 4.7769, 10.4167, 0.0208, C.ink);
const headerOpts = {
face: F.sans, size: 14.25, color: C.meta, charSpacing: 1.5, bold: true,
};
label(s, "MODEL", colX.model, headerY, colW.model, 0.25, headerOpts);
label(s, "ORG", colX.org, headerY, colW.org, 0.25, headerOpts);
label(s, "RELEASED", colX.rel, headerY, colW.rel, 0.25, headerOpts);
label(s, "SIGNIFICANCE", colX.sig, headerY, colW.sig, 0.25, headerOpts);
// Five rows: [ model, org, released, significance, rowY ]
const rows = [
["LLaMA", "Meta", "Feb 2023", "leaked in days; kicked off the open era", 5.0269, 5.6491],
["LLaMA 2", "Meta", "Jul 2023", "released commercially-usable weights", 5.8852, 6.5074],
["Mistral 7B", "Mistral", "Sep 2023", "matched 13B baselines", 6.7435, 7.3657],
["LLaMA 3", "Meta", "Apr 2024", "open frontier-class weights", 7.6018, 8.224 ],
["DeepSeek-V3 / R1", "DeepSeek", "2024–25", "open reasoning, low training cost", 8.4601, 9.4531],
];
for (const [model, org, rel, sig, y, ruleY] of rows) {
// Model — bold, brown
label(s, model, colX.model, y, colW.model, 0.42, {
face: F.sans, size: 19.5, color: C.ink, bold: true,
});
// Org — serif italic gold? No, XML shows just Helvetica
label(s, org, colX.org, y + 0.0764, colW.org, 0.3333, {
face: F.sans, size: 17, color: C.body,
});
// Released — mono meta
label(s, rel, colX.rel, y + 0.118, colW.rel, 0.2986, {
face: F.mono, size: 15, color: C.meta,
});
// Significance — body
label(s, sig, colX.sig, y + 0.0556, colW.sig, 0.3791, {
face: F.sans, size: 17, color: C.body,
});
// Thin rule below the row
bar(s, 8.3333, ruleY, 10.4167, 0.0104, C.rule);
}
}
// ─────────────────────────────────────────────────────────────────────────────
// SLIDE 8 — Modalities: 5-column grid (Text / Image / Audio / Video / Action)
// ─────────────────────────────────────────────────────────────────────────────
function buildSlide8() {
const s = pres.addSlide();
s.background = { color: C.bg };
addHeader(s, "CHAPTER VII · MODALITIES", "08 / 11");
addHeadline(s, "Models learned to see, hear, and speak.");
addBody(s,
"By 2024, the input stream widened: the same transformer backbone " +
"accepts tokens from text, images, audio, and video — and emits " +
"actions, not just words.",
1.25, 3.9212, 11.8021, 1.582
);
// 5 cards, only the first has a panel fill
const cards = [
{ x: 1.25, filled: true, glyph: "Aa", title: "Text", caption: "the original modality" },
{ x: 4.8083, filled: false, glyph: "◨", title: "Image", caption: "GPT-4V · Gemini · Claude 3" },
{ x: 8.3666, filled: false, glyph: "≋", title: "Audio", caption: "Whisper · voice-native models" },
{ x: 11.9249, filled: false, glyph: "▶", title: "Video", caption: "Sora, Veo — generation & understanding" },
{ x: 15.4833, filled: false, glyph: "⇢", title: "Action", caption: "tool use, code, UI control" },
];
for (const c of cards) {
if (c.filled) {
bar(s, c.x, 6.2116, 3.2666, 3.5417, C.panel);
}
// Giant glyph
label(s, c.glyph, c.x + 0.3056, 6.6838, 2.7388, 1.5, {
face: F.serif, size: 96, color: C.ink, align: "center", italic: true,
});
// Title
label(s, c.title, c.x + 0.3056, 8.4693, 2.7388, 0.4514, {
face: F.serif, size: 27, color: C.ink, align: "center",
});
// Caption
label(s, c.caption, c.x + 0.3056, 9.004, 2.7388, 0.5957, {
face: F.mono, size: 13.5, color: C.meta, align: "center",
});
}
}
// ─────────────────────────────────────────────────────────────────────────────
// SLIDE 9 — Reasoning: "thinking" panel with worked chain-of-thought
// ─────────────────────────────────────────────────────────────────────────────
function buildSlide9() {
const s = pres.addSlide();
s.background = { color: C.bg };
addHeader(s, "CHAPTER VIII · REASONING", "09 / 11");
addHeadline(s, "Reasoning became a training objective.");
addBody(s,
"Chain-of-thought prompting (2022) showed that models think better " +
"when allowed to write out intermediate steps.",
1.25, 4.561, 8.476, 1.582
);
addBody(s,
"OpenAI's o1 (2024) and DeepSeek-R1 (2025) made this a first-class " +
"training objective: reinforcement learning on long internal reasoning " +
"traces, with test-time compute traded for accuracy.",
1.25, 6.4347, 8.476, 2.0955
);
label(s, "TRAIN-TIME SCALE → TEST-TIME SCALE", 1.25, 8.9885, 8.476, 0.375, {
face: F.mono, size: 18, color: C.gold, charSpacing: 2.25,
});
// Right "THINKING" panel
bar(s, 10.5208, 4.7339, 8.2292, 4.4149, C.panel);
bar(s, 10.5208, 4.7339, 0.0417, 4.4149, C.goldHi);
label(s, "── THINKING ──", 11.0208, 5.1923, 7.489, 0.2778, {
face: F.mono, size: 16.5, color: C.meta, charSpacing: 1.5,
});
// Long chain-of-thought body (serif italic, slightly lighter)
s.addText(
"The user wants the second-largest prime factor. First, factor 2,024. " +
"It's even, so 2 × 1,012. Again, 2 × 506. Again, 2 × 253. 253 = 11 × 23. " +
"So primes are {2, 11, 23}. Second-largest is 11.",
{
x: 11.0208, y: 5.6784, w: 7.489, h: 2.4635,
fontFace: F.serif, italic: true, fontSize: 22, color: C.ink,
lineSpacingMultiple: 1.4, margin: 0.03, valign: "top",
}
);
// → answer: 11 (gold 11)
s.addText(
[
{ text: "→ answer: ", options: { color: C.meta, fontFace: F.mono, fontSize: 17 } },
{ text: "11", options: { color: C.gold, fontFace: F.serif, fontSize: 22, italic: true, bold: true } },
],
{ x: 11.0208, y: 8.4336, w: 7.489, h: 0.2986, margin: 0.03, valign: "middle" }
);
}
// ─────────────────────────────────────────────────────────────────────────────
// SLIDE 10 — Agency: the agent loop (illustration-driven)
// ─────────────────────────────────────────────────────────────────────────────
function buildSlide10() {
const s = pres.addSlide();
s.background = { color: C.bg };
addHeader(s, "CHAPTER IX · AGENCY", "10 / 11");
addHeadline(s, "Agents turn language models into actors.");
addBody(s,
"A model that can call tools, observe results, and iterate is an agent. " +
"Tool use (2023), function calling, and browser / computer control " +
"(2024–25) extend the output alphabet from words to actions.",
1.25, 4.7209, 7.725, 2.6089
);
addBody(s,
'The research frontier is no longer "what can a model answer?" but ' +
'"what can a model accomplish, unsupervised, over long horizons?"',
1.25, 7.6215, 7.725, 1.582
);
// Illustration
if (hasAsset("image-10-1.png")) {
s.addImage({ path: asset("image-10-1.png"), x: 9.5833, y: 3.4661, w: 9.1667, h: 6.9505 });
}
}
// ─────────────────────────────────────────────────────────────────────────────
// SLIDE 11 — Coda: 2x2 grid of open questions
// ─────────────────────────────────────────────────────────────────────────────
function buildSlide11() {
const s = pres.addSlide();
s.background = { color: C.bg };
addHeader(s, "CODA", "11 / 11");
addHeadline(s, "Four open questions for the next decade.");
const items = [
{ x: 1.25, y: 4.7839, labelX: 1.25, textX: 2.375, roman: "i .", q: "How far does pre-training scale?", a: "data is finite; frontier runs cost $100M+", qW: 6.0153, aW: 6.0153 },
{ x: 10.3125, y: 4.7839, labelX: 10.3125, textX: 11.4375, roman: "ii .", q: "What's the right test-time compute curve?", a: "o-series suggests reasoning is elastic", qW: 7.5319, aW: 7.5319 },
{ x: 1.25, y: 7.2296, labelX: 1.25, textX: 2.375, roman: "iii .", q: "Do agents reliably generalize?", a: "long horizons remain the bottleneck", qW: 5.5176, aW: 5.5176 },
{ x: 10.3125, y: 7.2296, labelX: 10.3125, textX: 11.4375, roman: "iv .", q: "Where do the weights live?", a: "open vs closed, nation-state vs lab", qW: 4.971, aW: 4.971 },
];
for (const it of items) {
// Thin top rule
bar(s, it.x, it.y, 8.4375, 0.0208, C.ink);
// Roman numeral (serif italic gold)
label(s, it.roman, it.labelX, it.y + 0.3125, 0.9167, 1.05, {
face: F.serif, size: 27, color: C.gold, italic: true,
});
// Question (serif, brown). Drop from 25.5 pt to 22 pt so the longer
// questions (ii.) stay on a single line.
label(s, it.q, it.textX, it.y + 0.3125, it.qW, 0.75, {
face: F.serif, size: 22, color: C.ink,
});
// Answer (sans, muted). Nudge down so there's a clean gap below the question.
label(s, it.a, it.textX, it.y + 1.05, it.aW, 0.4041, {
face: F.sans, size: 14, color: C.body,
});
}
// Footer rule + "Thank you · Questions?"
bar(s, 1.25, 9.6597, 17.5, 0.0104, C.rule);
bar(s, 1.25, 10.2344, 0.625, 0.0312, C.goldHi);
label(s, "Thank you · Questions?", 2.1667, 10.0833, 4.2305, 0.375, {
face: F.mono, size: 18, color: C.meta, charSpacing: 1.5,
});
}
// ─────────────────────────────────────────────────────────────────────────────
// Build everything
// ─────────────────────────────────────────────────────────────────────────────
buildSlide1();
buildSlide2();
buildSlide3();
buildSlide4();
buildSlide5();
buildSlide6();
buildSlide7();
buildSlide8();
buildSlide9();
buildSlide10();
buildSlide11();
// ─────────────────────────────────────────────────────────────────────────────
// Write
// ─────────────────────────────────────────────────────────────────────────────
pres.writeFile({ fileName: "class2-replica.pptx" }).then((f) => {
console.log("Wrote:", f);
});
Related skills
| Skill | Stars | Updated | Works with | Status | Install |
|---|---|---|---|---|---|
| canvas-designanthropics/skillsCreate beautiful visual art in .png and .pdf documents using design philosophy. You should use this skill when the user asks to create a poster, piece of art, design, or other static piece. Create original visual designs, never copying existing artists' work to avoid copyright violations. | Stars: 198k | Updated | Claude CodeCodex | Verified | Install |
| docxanthropics/skillsUse this skill whenever the user wants to create, read, edit, or manipulate Word documents (.docx files) or Word templates (.dotx files). Triggers include: any mention of 'Word doc', 'word document', '.docx', '.dotx', or requests to produce professional documents with formatting like tables of contents, headings, page numbers, or letterheads. Also use when extracting or reorganizing content from .docx or .dotx files, inserting or replacing images in documents, performing find-and-replace in Word files, working with tracked changes or comments, or converting content into a polished Word document. If the user asks for a 'report', 'memo', 'letter', 'template', or similar deliverable as a Word or .docx file, use this skill. Do NOT use for PDFs, spreadsheets, Google Docs, or general coding tasks unrelated to document generation. | Stars: 198k | Updated | Claude CodeCodex | Verified | Install |
| pdfanthropics/skillsUse this skill whenever the user wants to do anything with PDF files. This includes reading or extracting text/tables from PDFs, combining or merging multiple PDFs into one, splitting PDFs apart, rotating pages, adding watermarks, creating new PDFs, filling PDF forms, encrypting/decrypting PDFs, extracting images, and OCR on scanned PDFs to make them searchable. If the user mentions a .pdf file or asks to produce one, use this skill. | Stars: 198k | Updated | Claude CodeCodex | Verified | Install |
| pptxanthropics/skillsUse this skill any time a .pptx or .potx file is involved in any way — as input, output, or both. This includes: creating slide decks, pitch decks, or presentations; reading, parsing, or extracting text from any .pptx or .potx file (even if the extracted content will be used elsewhere, like in an email or summary); editing, modifying, or updating existing presentations; combining or splitting slide files; working with templates (.potx), layouts, speaker notes, or comments. Trigger whenever the user mentions "deck," "slides," "presentation," or references a .pptx or .potx filename, regardless of what they plan to do with the content afterward. If a .pptx or .potx file needs to be opened, created, or touched, use this skill. | Stars: 198k | Updated | Claude CodeCodex | Verified | Install |