27:null 28:["$","title",null,{"children":"Efficient Frontier of Compute | North Branch Media | Observable"}] 2d:T6f9d,// Efficient Frontier of Compute // Sources found here https://github.com/NorthBranchMedia/The-Efficient-Frontier-of-Compute { const d3 = await require("d3@7"); // Theme const NB_BLUE = "#043463"; const NB_ORANGE = "#f57c00"; const BG = NB_BLUE; const TEXT_PRIMARY = "#ffffff"; const TEXT_SECONDARY = "#ffffff"; const AXIS_LINE = "rgba(255,255,255,0.55)"; const FRONTIER_LINE = "#ffffff"; const LANDAUER_LINE = "rgba(255,200,120,0.9)"; const LEADER_LINE = "rgba(255,255,255,0.4)"; const LABEL_HALO = NB_BLUE; const DOT_STROKE = "rgba(255,255,255,0.7)"; const STATUS_TEXT_COLOR = { shipping: "#ff6b6b", prototype: "#ffb3b3", theoretical: "#cccccc" }; const STATUS_LABEL = { shipping: "Commercial", prototype: "Researching", theoretical: "Theoretical" }; const STATUS_FILL = { shipping: "#fe0100", prototype: "#ff8088", theoretical: "#ffffff" }; // LLM Cloud const TAX_COLORS = { brain: { fill: "rgba(233,30,99,0.125)", stroke: "rgba(233,30,99,0.9)", label: "Human Brain" }, llm: { fill: "rgba(120,170,255,0.125)", stroke: "rgba(150,190,255,0.95)", label: "Frontier LLM" } }; const CLOUDS_TO_RENDER = new Set(["brain", "llm"]); const CLOUD_CENTROID_LABEL = { brain: "Human Brain", llm: "Frontier LLM" }; const MANUAL_LABELS = new Set(["Brain: short reply", "Brain: chess", "Brain: book", "Brain: AIME", "LLM: short reply", "LLM: chess", "LLM: book", "LLM: AIME", "Early-2020s LLM", "Ribosome", "E. coli", "Honeybee", "Frog", "Dolphin", "ENIAC", "Loihi 2", "TrueNorth", "Extropic TSU", "Antikythera", "Lightmatter Envise", "Willow", "Google Search"]); const raw = [ { tax: "brain", label: "Brain: short reply", hw: "Organic", software: "Learned Weights", firstYear: "~300 ka", tier: "T2 sourced envelope", task: "Produce ~110-word reply (reads prompt at 238 wpm, speaks at 150 wpm)", sec: 69.2101, joules: 1707.22, bits: 652.96, status: "shipping", labelDx: 8, labelDy: 60 }, { tax: "brain", label: "Brain: chess", hw: "Organic", software: "Learned Weights", firstYear: "~300 ka", tier: "T2 sourced envelope", task: "Choose best move in a sharp position (~4 candidates = 2 bits), 180 s classical control", sec: 180, joules: 4440.09, bits: 2, status: "shipping", labelDx: 16, labelDy: 90 }, { tax: "brain", label: "Brain: book", hw: "Organic", software: "Learned Weights", firstYear: "~300 ka", tier: "T2 sourced envelope", task: "Read 80k-word book, one-word verdict. INPUT-BIT BASIS (exception): 474,880 bits ingested", sec: 20168.07, joules: 497489, bits: 474880, status: "shipping", labelDx: -90, labelDy: 65 }, { tax: "brain", label: "Brain: AIME", hw: "Organic", software: "Learned Weights", firstYear: "~300 ka", tier: "T2 sourced envelope", task: "Solve one AIME problem (answer 000-999 = 9.97 bits), 720 s allowance", sec: 720, joules: 17760.4, bits: 9.9658, status: "shipping", labelDx: -40, labelDy: 116 }, { tax: "llm", label: "LLM: short reply", hw: "Charge", software: "Universal", firstYear: 2025, tier: "T2 modelled", task: "Short reply, 278.25 tokens at 2.10 J/token", sec: 0.5565, joules: 584.325, bits: 652.96, status: "shipping", labelDx: -90, labelDy: 95 }, { tax: "llm", label: "LLM: chess", hw: "Charge", software: "Universal", firstYear: 2025, tier: "T3 reasoning est.", task: "Choose best move, ~885 tokens (incl. ~800 reasoning)", sec: 1.769, joules: 1857.45, bits: 2, status: "shipping", labelDx: -50, labelDy: -34 }, { tax: "llm", label: "LLM: book", hw: "Charge", software: "Universal", firstYear: 2025, tier: "T2 modelled", task: "Read 80k-word book. INPUT-BIT BASIS (exception); Epoch 40 Wh", sec: 137.143, joules: 144000, bits: 474880, status: "shipping", labelDx: -100, labelDy: -60 }, { tax: "llm", label: "LLM: AIME", hw: "Charge", software: "Universal", firstYear: 2025, tier: "T2 modelled", task: "Solve one AIME problem, ~10,085 tokens (incl. 10k reasoning)", sec: 20.169, joules: 21177.5, bits: 9.9658, status: "shipping", labelDx: 44, labelDy: -28 }, { tax: "charge", label: "Early-2020s LLM", hw: "Charge", software: "Universal", firstYear: 2021, tier: "T3 reconstruction", task: "Produce ~110-word reply (GPT-3-class, A100, x1.5 facility)", sec: 1.04046, joules: 624.279, bits: 652.96, status: "shipping", labelDx: -55, labelDy: -70 }, { tax: "organic", label: "Ribosome", hw: "Organic", software: "Fixed", firstYear: "~3.8 Gya", tier: "T2 modelled", task: "Synthesise one 300-residue protein (wide energy boundary, 4.19 bits/residue)", sec: 15, joules: 4.482e-16, bits: 1257, status: "shipping", labelDx: 44, labelDy: -14 }, { tax: "organic", label: "E. coli", hw: "Organic", software: "Fixed", firstYear: "~3.5 Gya", tier: "T2 sourced", task: "One run/tumble chemotaxis commitment (whole-cell)", sec: 1, joules: 1e-12, bits: 1, status: "shipping", labelDx: 26, labelDy: -24 }, { tax: "organic", label: "Honeybee", hw: "Organic", software: "Learned Weights", firstYear: "~100 Ma", tier: "T3 reconstruction", task: "Select 1 of ~8 remembered forage patches", sec: 0.2, joules: 6.72e-5, bits: 3, status: "shipping", labelDx: 50, labelDy: -20 }, { tax: "organic", label: "Frog", hw: "Organic", software: "Learned Weights", firstYear: "~200 Ma", tier: "T4 reconstruction", task: "Detect prey, commit to strike direction (1 of 8). Ectotherm", sec: 0.15, joules: 1.5e-4, bits: 3, status: "shipping", labelDx: -95, labelDy: 50 }, { tax: "organic", label: "Dolphin", hw: "Organic", software: "Learned Weights", firstYear: "~50 Ma", tier: "T3 method conflict (6x)", task: "One 2AFC echolocation discrimination", sec: 2, joules: 21.2, bits: 1, status: "shipping", labelDx: 12, labelDy: 40 }, { tax: "charge", label: "ENIAC", hw: "Charge", software: "Universal", firstYear: 1945, tier: "T1 sourced", task: "One 10-digit decimal addition (150 kW x1.4 est. cooling)", sec: 2e-4, joules: 42, bits: 33.22, status: "shipping", labelDx: -80, labelDy: 0 }, { tax: "charge", label: "Loihi 2", hw: "Charge", software: "Learned Weights", firstYear: 2021, tier: "T2 sourced", task: "One synaptic operation (x1.5 facility)", sec: 1e-5, joules: 2.25e-11, bits: 1, status: "shipping", labelDx: 54, labelDy: -16 }, { tax: "charge", label: "TrueNorth", hw: "Charge", software: "Learned Weights", firstYear: 2014, tier: "T1 sourced", task: "One synaptic event (x1.5 facility)", sec: 1e-3, joules: 3.9e-11, bits: 1, status: "shipping", labelDx: 44, labelDy: -20 }, { tax: "charge", label: "Extropic TSU", hw: "Charge", software: "Probabilistic", firstYear: 2025, tier: "T4 vendor claim - simulated", task: "One probabilistic sample (simulated, x1.5 facility)", sec: 1e-1, joules: 1.5e-5, bits: 1, status: "prototype", labelDx: 0, labelDy: 60 }, { tax: "kinetic", label: "Antikythera", hw: "Kinetic", software: "Fixed", firstYear: "~100 BC", tier: "T4 guess - muscle input", task: "One eclipse/position dial readout (operator crank effort). Output-width bits", sec: 1e1, joules: 1e1, bits: 15, status: "shipping", labelDx: 70, labelDy: 100 }, { tax: "wave", label: "Lightmatter Envise", hw: "Wave", software: "Fixed", firstYear: 2021, tier: "T4 vendor claim", task: "One 8-bit multiply-accumulate (x1.5 facility). Output-width bits, not a decision", sec: 1e-14, joules: 7.5e-12, bits: 8, status: "shipping", labelDx: 10, labelDy: -20 }, { tax: "superconductor", label: "Willow", hw: "Superconductor", software: "Superposition", firstYear: 2024, tier: "T3 reconstruction - Nature gives no energy", task: "One 105-qubit RCS sample (full cryoplant)", sec: 3e-3, joules: 45, bits: 105, status: "prototype", labelDx: -90, labelDy: 30 }, { tax: "charge", label: "Google Search", hw: "Charge", software: "Universal", firstYear: 2009, tier: "T2 modelled - source stale", task: "Return one ranked result (2009 disclosure, fleet-wide)", sec: 0.2, joules: 1080, bits: 36.54, status: "shipping", labelDx: -30, labelDy: -60 } ]; const points = raw.map(p => ({ ...p, x: Math.log10(p.sec / p.bits), y: Math.log10(p.joules / p.bits) })); const totalWidth = 1200, totalHeight = 800; const margin = { top: 90, right: 90, bottom: 90, left: 90 }; const innerW = totalWidth - margin.left - margin.right; const innerH = totalHeight - margin.top - margin.bottom; const xExt = d3.extent(points, d => d.x); const yExt = d3.extent(points, d => d.y); const xDom = [Math.floor(xExt[0]) - 1, Math.ceil(xExt[1]) + 1]; const yDom = [Math.ceil(yExt[1]) + 1, Math.floor(yExt[0]) - 1]; const landauerLog = Math.log10(2.87e-21); if (yDom[1] > landauerLog) yDom[1] = Math.floor(landauerLog) - 1; const xScale = d3.scaleLinear().domain(xDom).range([0, innerW]); const yScale = d3.scaleLinear().domain(yDom).range([0, innerH]); const wrap = d3.create("div").style("position", "relative").style("width", "100%").style("line-height", "0").style("font-size", "0"); const svg = wrap.append("svg") .attr("viewBox", [0, 0, totalWidth, totalHeight]) .attr("preserveAspectRatio", "xMidYMid meet") .attr("width", "100%").style("display", "block").style("background", BG) .style("font-family", "'Helvetica Neue', Arial, sans-serif"); const root = svg.append("g").attr("transform", `translate(${margin.left},${margin.top})`); // Gridlines const GRID_COLOR = "rgba(255,255,255,0.35)"; const gridLayer = root.append("g").attr("class", "grid"); for (let t = xDom[0]; t <= xDom[1]; t += 1) { gridLayer.append("line").attr("x1", xScale(t)).attr("x2", xScale(t)).attr("y1", 0).attr("y2", innerH) .attr("stroke", GRID_COLOR).attr("stroke-width", 0.5).attr("stroke-dasharray", "1 3"); } for (let t = yDom[1]; t <= yDom[0]; t += 1) { gridLayer.append("line").attr("x1", 0).attr("x2", innerW).attr("y1", yScale(t)).attr("y2", yScale(t)) .attr("stroke", GRID_COLOR).attr("stroke-width", 0.5).attr("stroke-dasharray", "1 3"); } const tt = wrap.append("div") .style("position", "absolute").style("background", "rgba(255,255,255,0.98)") .style("border", "1px solid #bbb").style("border-radius", "6px").style("padding", "10px 14px") .style("font-family", "'Helvetica Neue',Arial,sans-serif").style("font-size", "13px").style("color", "#222") .style("line-height", "1.65").style("pointer-events", "none").style("box-shadow", "0 2px 12px rgba(0,0,0,.18)") .style("display", "none").style("max-width", "300px").style("z-index", "20"); function showTT(ev, html) { const [mx, my] = d3.pointer(ev, wrap.node()); const ww = wrap.node().getBoundingClientRect().width; const left = mx + 16 + 300 > ww ? mx - 316 : mx + 16; tt.style("display", "block").html(html).style("left", left + "px").style("top", Math.max(0, my - 12) + "px"); } function hideTT() { tt.style("display", "none"); } const SUP = {'-':'⁻','0':'⁰','1':'¹','2':'²','3':'³','4':'⁴','5':'⁵','6':'⁶','7':'⁷','8':'⁸','9':'⁹'}; const toSuper = (n) => String(n).split('').map(c => SUP[c] || c).join(''); function fmtNum(v) { if (v === 0) return "0"; const abs = Math.abs(v); if (abs >= 0.01 && abs < 10000) { if (Number.isInteger(v)) return String(v); return String(+v.toPrecision(3)); } const exp = Math.floor(Math.log10(abs)); const m = v / Math.pow(10, exp); const mStr = Math.abs(m - 1) < 1e-9 ? "10" : `${+m.toPrecision(2)}×10`; return mStr + String(exp).split('').map(c => SUP[c] || c).join(''); } const TIER_COLOR = (t) => t.startsWith("T1") ? "#2e7d32" : t.startsWith("T2") ? "#689f38" : t.startsWith("T3") ? "#ef6c00" : "#c62828"; function pointHTML(p) { const jpb = p.joules / p.bits, spb = p.sec / p.bits, overL = jpb / 2.87e-21; return `

${p.label}

` + `

${STATUS_LABEL[p.status]}

` + `

${p.tier}

` + `

${p.task}

` + `

Hardware: ${p.hw}

Software: ${p.software}

` + `

First Demonstrated: ${p.firstYear}

` + `

Time: ${fmtNum(p.sec)} s

Energy: ${fmtNum(p.joules)} J

` + `

Bits resolved: ${fmtNum(p.bits)}

` + `

${fmtNum(jpb)} J / bit

${fmtNum(spb)} s / bit

` + `

${fmtNum(overL)}× above Landauer

`; } function convexHull(pts) { if (pts.length < 3) return pts.slice(); const sorted = pts.slice().sort((a, b) => a[0] - b[0] || a[1] - b[1]); const cross = (O, A, B) => (A[0] - O[0]) * (B[1] - O[1]) - (A[1] - O[1]) * (B[0] - O[0]); const lower = []; for (const p of sorted) { while (lower.length >= 2 && cross(lower[lower.length - 2], lower[lower.length - 1], p) <= 0) lower.pop(); lower.push(p); } const upper = []; for (let i = sorted.length - 1; i >= 0; i--) { const p = sorted[i]; while (upper.length >= 2 && cross(upper[upper.length - 2], upper[upper.length - 1], p) <= 0) upper.pop(); upper.push(p); } upper.pop(); lower.pop(); return lower.concat(upper); } function fluffPoints(pts, radius, segs) { const result = []; const n = segs || 24; for (const [px, py] of pts) for (let i = 0; i < n; i++) { const a = (i / n) * Math.PI * 2; result.push([px + Math.cos(a) * radius, py + Math.sin(a) * radius]); } return result; } const byTax = d3.group(points, d => d.tax); const cloudLayer = root.append("g").attr("class", "clouds"); const cloudCentroids = []; for (const [tax, taxPoints] of byTax) { if (!CLOUDS_TO_RENDER.has(tax)) continue; const pixelPts = taxPoints.map(p => [xScale(p.x), yScale(p.y)]); if (pixelPts.length < 2) continue; const fluffed = fluffPoints(pixelPts, 26, 24); const hull = convexHull(fluffed); if (hull.length < 3) continue; const line = d3.line().curve(d3.curveCatmullRomClosed.alpha(0.5)); cloudLayer.append("path").attr("d", line(hull)).attr("fill", TAX_COLORS[tax].fill).attr("stroke", TAX_COLORS[tax].stroke).attr("stroke-width", 1.2); cloudCentroids.push({ tax, cx: d3.mean(pixelPts, d => d[0]), cy: d3.mean(pixelPts, d => d[1]) }); } if (landauerLog >= yDom[1] && landauerLog <= yDom[0]) { root.append("line").attr("x1", 0).attr("x2", innerW).attr("y1", yScale(landauerLog)).attr("y2", yScale(landauerLog)) .attr("stroke", LANDAUER_LINE).attr("stroke-width", 1.5).attr("stroke-dasharray", "6 4"); root.append("text").attr("x", innerW - 8).attr("y", yScale(landauerLog) - 6).attr("fill", LANDAUER_LINE) .attr("font-size", 12).attr("font-style", "italic").attr("text-anchor", "end").text("Landauer limit: kT·ln(2) at 300 K"); } const frontierPool = points.filter(p => p.status !== "theoretical"); const pareto = frontierPool.filter(p => frontierPool.every(q => q === p || !(q.x <= p.x && q.y < p.y))).sort((a, b) => a.x - b.x); let frontierLabelRect = null; if (pareto.length >= 2) { const stair = []; stair.push([pareto[0].x, pareto[0].y]); for (let i = 1; i < pareto.length; i++) { stair.push([pareto[i].x, pareto[i - 1].y]); stair.push([pareto[i].x, pareto[i].y]); } const frontierLine = d3.line().x(d => xScale(d[0])).y(d => yScale(d[1])).curve(d3.curveLinear); root.append("path").datum(stair).attr("d", frontierLine).attr("fill", "none").attr("stroke", FRONTIER_LINE).attr("stroke-width", 1.4).attr("stroke-dasharray", "2 3"); const flXLog = -9, flYLog = -11.5; const flX = xScale(flXLog), flY = yScale(flYLog); root.append("text").attr("x", flX).attr("y", flY).attr("fill", FRONTIER_LINE).attr("font-size", 12).attr("font-style", "italic").text("Efficient Frontier"); const probe = svg.append("text").attr("opacity", 0).attr("font-size", 12).attr("font-style", "italic").text("Efficient Frontier"); let fw; try { fw = probe.node().getComputedTextLength(); } catch (e) { fw = 110; } probe.remove(); frontierLabelRect = { x: flX - 2, y: flY - 11, w: fw + 4, h: 14 }; } const axisLayer = root.append("g").attr("class", "axes"); axisLayer.append("line").attr("x1", 0).attr("x2", innerW).attr("y1", innerH).attr("y2", innerH).attr("stroke", AXIS_LINE).attr("stroke-width", 1); for (let t = xDom[0]; t <= xDom[1]; t += 2) axisLayer.append("text").attr("x", xScale(t)).attr("y", innerH + 18).attr("fill", TEXT_SECONDARY).attr("font-size", 12).attr("text-anchor", "middle").text(`10${toSuper(t)}`); axisLayer.append("text").attr("x", innerW / 2).attr("y", innerH + 52).attr("fill", TEXT_PRIMARY).attr("font-size", 14).attr("font-weight", 600).attr("text-anchor", "middle").text("Seconds per Bit Resolved (log)"); axisLayer.append("line").attr("x1", 0).attr("x2", 0).attr("y1", 0).attr("y2", innerH).attr("stroke", AXIS_LINE).attr("stroke-width", 1); for (let t = yDom[1]; t <= yDom[0]; t += 2) axisLayer.append("text").attr("x", -10).attr("y", yScale(t) + 4).attr("fill", TEXT_SECONDARY).attr("font-size", 12).attr("text-anchor", "end").text(`10${toSuper(t)}`); axisLayer.append("text").attr("transform", `rotate(-90)`).attr("x", -innerH / 2).attr("y", -56).attr("fill", TEXT_PRIMARY).attr("font-size", 14).attr("font-weight", 600).attr("text-anchor", "middle").text("Joules per Bit Resolved (log)"); const labelled = points.slice(); const labelDims = labelled.map(p => { const probe = svg.append("text").attr("opacity", 0).attr("font-size", 13.2).text(p.label); let w; try { w = probe.node().getComputedTextLength(); } catch (e) { w = p.label.length * 7.44; } probe.remove(); return { width: w + 6, height: 16 }; }); const DOT_R = 9; const dotRects = points.map(p => ({ x: xScale(p.x) - DOT_R, y: yScale(p.y) - DOT_R, w: DOT_R * 2, h: DOT_R * 2 })); const reservedRects = []; { const lText = "Landauer limit: kT·ln(2) at 300 K"; const probe = svg.append("text").attr("opacity", 0).attr("font-size", 12).attr("font-style", "italic").text(lText); let lw; try { lw = probe.node().getComputedTextLength(); } catch (e) { lw = lText.length * 6.4; } probe.remove(); reservedRects.push({ x: innerW - 8 - lw, y: yScale(landauerLog) - 18, w: lw + 4, h: 16 }); } if (frontierLabelRect) reservedRects.push(frontierLabelRect); for (const c of cloudCentroids) reservedRects.push({ x: c.cx - 52, y: c.cy - 11, w: 104, h: 22 }); function rectsOverlap(a, b) { return !(a.x + a.w <= b.x || b.x + b.w <= a.x || a.y + a.h <= b.y || b.y + b.h <= a.y); } function overlapArea(a, b) { const xo = Math.max(0, Math.min(a.x+a.w, b.x+b.w) - Math.max(a.x, b.x)); const yo = Math.max(0, Math.min(a.y+a.h, b.y+b.h) - Math.max(a.y, b.y)); return xo * yo; } function labelRectFor(dotX, dotY, dx, dy, lw, lh) { return { x: dotX + dx - lw/2, y: dotY + dy + 3 - 8, w: lw, h: lh }; } function generateCandidates(lw, lh, preferredDx, preferredDy) { const cands = []; const safety = 4; { const r = labelRectFor(0, 0, preferredDx, preferredDy, lw, lh); const ownDot = { x: -DOT_R, y: -DOT_R, w: DOT_R*2, h: DOT_R*2 }; if (!rectsOverlap(r, ownDot)) cands.push({ dx: preferredDx, dy: preferredDy, preferred: true }); } const angles = 36; const offsets = [0, 8, 18, 32, 52, 78, 112, 155, 210]; for (let i = 0; i < angles; i++) { const a = (i / angles) * Math.PI * 2, cosA = Math.cos(a), sinA = Math.sin(a); const minRX = Math.abs(cosA) > 0.05 ? (lw/2 + DOT_R + safety) / Math.abs(cosA) : Infinity; const minRY = Math.abs(sinA) > 0.05 ? (lh/2 + DOT_R + safety) / Math.abs(sinA) : Infinity; const minR = Math.min(minRX, minRY); if (!isFinite(minR)) continue; for (const off of offsets) { const r = minR + off; cands.push({ dx: cosA * r, dy: sinA * r }); } } return cands; } const densityR = 80; const placementOrder = labelled.map((p, i) => { const px = xScale(p.x), py = yScale(p.y); let n = 0; for (const q of points) { if (q === p) continue; if (Math.hypot(xScale(q.x) - px, yScale(q.y) - py) < densityR) n++; } return { i, n }; }).sort((a, b) => b.n - a.n).map(o => o.i); const boundsXMin = -margin.left + 5, boundsXMax = innerW + margin.right - 5, boundsYMin = -margin.top + 55, boundsYMax = innerH + margin.bottom - 55; const placedRects = [...reservedRects]; const finalPlacements = new Array(labelled.length); for (const idx of placementOrder) { const p = labelled[idx]; if (!MANUAL_LABELS.has(p.label)) continue; const { width: lw, height: lh } = labelDims[idx]; const dotX = xScale(p.x), dotY = yScale(p.y); finalPlacements[idx] = { dx: p.labelDx, dy: p.labelDy }; placedRects.push(labelRectFor(dotX, dotY, p.labelDx, p.labelDy, lw, lh)); } for (const idx of placementOrder) { const p = labelled[idx]; if (MANUAL_LABELS.has(p.label)) continue; const { width: lw, height: lh } = labelDims[idx]; const dotX = xScale(p.x), dotY = yScale(p.y); let bestClean = null, bestCleanScore = Infinity, bestDirty = null, bestDirtyOverlap = Infinity; const cands = generateCandidates(lw, lh, p.labelDx, p.labelDy); for (const c of cands) { const rect = labelRectFor(dotX, dotY, c.dx, c.dy, lw, lh); if (rect.x < boundsXMin || rect.y < boundsYMin || rect.x + rect.w > boundsXMax || rect.y + rect.h > boundsYMax) continue; let collides = false, totalOverlap = 0; for (let i = 0; i < dotRects.length; i++) { if (points[i] === p) continue; if (rectsOverlap(rect, dotRects[i])) { collides = true; totalOverlap += overlapArea(rect, dotRects[i]); } } for (const pr of placedRects) { if (rectsOverlap(rect, pr)) { collides = true; totalOverlap += overlapArea(rect, pr); } } const distFromDot = Math.hypot(c.dx, c.dy), distFromPref = Math.hypot(c.dx - p.labelDx, c.dy - p.labelDy); if (!collides) { const prefBonus = c.preferred ? -60 : 0; const score = distFromDot + distFromPref * 0.25 + prefBonus; if (score < bestCleanScore) { bestCleanScore = score; bestClean = { dx: c.dx, dy: c.dy, rect }; } } else if (!bestClean && totalOverlap < bestDirtyOverlap) { bestDirtyOverlap = totalOverlap; bestDirty = { dx: c.dx, dy: c.dy, rect }; } } const chosen = bestClean || bestDirty; if (chosen) { finalPlacements[idx] = { dx: chosen.dx, dy: chosen.dy }; placedRects.push(chosen.rect); } else { const cx = innerW / 2, cy = innerH / 2; const angle = Math.atan2(dotY - cy, dotX - cx); const fdx = Math.cos(angle) * 120, fdy = Math.sin(angle) * 120; finalPlacements[idx] = { dx: fdx, dy: fdy }; placedRects.push(labelRectFor(dotX, dotY, fdx, fdy, lw, lh)); } } labelled.forEach((p, idx) => { p.labelDx = finalPlacements[idx].dx; p.labelDy = finalPlacements[idx].dy; }); // Clean { const rectOf = (idx) => { const p = labelled[idx]; const { width: lw, height: lh } = labelDims[idx]; const dotX = xScale(p.x), dotY = yScale(p.y); return labelRectFor(dotX, dotY, finalPlacements[idx].dx, finalPlacements[idx].dy, lw, lh); }; const PAD = 4; const inflate = (r, m) => ({ x: r.x - m, y: r.y - m, w: r.w + 2*m, h: r.h + 2*m }); // static obstacles: all dots (small squares) + reserved rects const dotObstacles = points.map(p => ({ x: xScale(p.x) - 5, y: yScale(p.y) - 5, w: 10, h: 10 })); const staticObstacles = dotObstacles.concat(reservedRects); for (let iter = 0; iter < 600; iter++) { let moved = false; // label vs label for (let i = 0; i < labelled.length; i++) { for (let j = i + 1; j < labelled.length; j++) { const a = rectOf(i), b = rectOf(j); const ox = Math.min(a.x+a.w, b.x+b.w) - Math.max(a.x, b.x); const oy = Math.min(a.y+a.h, b.y+b.h) - Math.max(a.y, b.y); if (ox > -PAD && oy > -PAD) { moved = true; if (ox < oy) { const push = (ox + PAD)/2 + 0.5; const dir = (a.x < b.x) ? -1 : 1; finalPlacements[i].dx += dir*push; finalPlacements[j].dx -= dir*push; } else { const push = (oy + PAD)/2 + 0.5; const dir = (a.y < b.y) ? -1 : 1; finalPlacements[i].dy += dir*push; finalPlacements[j].dy -= dir*push; } } } } // Labels for (let i = 0; i < labelled.length; i++) { const a = rectOf(i); const ownDotX = xScale(labelled[i].x), ownDotY = yScale(labelled[i].y); for (const ob of staticObstacles) { // skip a label's own dot at very center (it will be pulled away anyway) const ox = Math.min(a.x+a.w, ob.x+ob.w) - Math.max(a.x, ob.x); const oy = Math.min(a.y+a.h, ob.y+ob.h) - Math.max(a.y, ob.y); if (ox > -PAD && oy > -PAD) { moved = true; if (ox < oy) { const push = (ox + PAD) + 0.5; const dir = ((a.x + a.w/2) < (ob.x + ob.w/2)) ? -1 : 1; finalPlacements[i].dx += dir*push; } else { const push = (oy + PAD) + 0.5; const dir = ((a.y + a.h/2) < (ob.y + ob.h/2)) ? -1 : 1; finalPlacements[i].dy += dir*push; } } } } if (!moved) break; } labelled.forEach((p, idx) => { p.labelDx = finalPlacements[idx].dx; p.labelDy = finalPlacements[idx].dy; }); } const leaderLayer = root.append("g").attr("class", "leaders"); labelled.forEach(p => { const dx = p.labelDx, dy = p.labelDy, dist = Math.sqrt(dx * dx + dy * dy); if (dist > 16) { const dotX = xScale(p.x), dotY = yScale(p.y), stopShort = 12, t = Math.max(0, (dist - stopShort) / dist); leaderLayer.append("line").attr("x1", dotX).attr("y1", dotY).attr("x2", dotX + dx * t).attr("y2", dotY + dy * t).attr("stroke", LEADER_LINE).attr("stroke-width", 0.8); } }); const pointLayer = root.append("g").attr("class", "points"); points.forEach(p => { const cx = xScale(p.x), cy = yScale(p.y), inCloud = CLOUDS_TO_RENDER.has(p.tax); pointLayer.append("circle").attr("cx", cx).attr("cy", cy).attr("r", inCloud ? 5 : 4.5) .attr("fill", inCloud ? TAX_COLORS[p.tax].stroke : STATUS_FILL[p.status]).attr("stroke", DOT_STROKE).attr("stroke-width", 0.8) .style("cursor", "pointer").on("mousemove", (ev) => showTT(ev, pointHTML(p))).on("mouseleave", hideTT); }); const labelLayer = root.append("g").attr("class", "labels"); labelled.forEach(p => { labelLayer.append("text").attr("x", xScale(p.x) + p.labelDx).attr("y", yScale(p.y) + p.labelDy + 3) .attr("fill", TEXT_PRIMARY).attr("font-size", 13.2).attr("text-anchor", "middle") .attr("paint-order", "stroke").attr("stroke", LABEL_HALO).attr("stroke-width", 2.8).attr("stroke-linejoin", "round") .style("pointer-events", "none").text(p.label); }); const taxLegend = svg.append("g").attr("transform", `translate(${margin.left},${margin.top - 50})`); let cursorX = 0; Object.entries(TAX_COLORS).forEach(([tax, c]) => { const itemG = taxLegend.append("g").attr("transform", `translate(${cursorX},0)`); itemG.append("circle").attr("cx", 7).attr("cy", 7).attr("r", 7).attr("fill", c.fill).attr("stroke", c.stroke).attr("stroke-width", 1.2); const txt = itemG.append("text").attr("x", 20).attr("y", 11).attr("fill", TEXT_PRIMARY).attr("font-size", 12).text(c.label); let textW = 0; try { textW = txt.node().getComputedTextLength(); } catch (e) {} if (!textW) textW = c.label.length * 6.5; cursorX += 20 + textW + 24; }); const statusLegend = svg.append("g"); const statusItems = [{ fill: STATUS_FILL.shipping, label: "Commercial" }, { fill: STATUS_FILL.prototype, label: "Researching" }]; let sX = 0, lastRightEdge = 0; statusItems.forEach(item => { const g = statusLegend.append("g").attr("transform", `translate(${sX},0)`); g.append("circle").attr("cx", 6).attr("cy", 7).attr("r", 4).attr("fill", item.fill).attr("stroke", DOT_STROKE).attr("stroke-width", 0.6); const t = g.append("text").attr("x", 18).attr("y", 11).attr("fill", TEXT_PRIMARY).attr("font-size", 12).text(item.label); let textW = 0; try { textW = t.node().getComputedTextLength(); } catch (e) {} if (!textW) textW = item.label.length * 6.5; lastRightEdge = sX + 18 + textW; sX += 18 + textW + 22; }); const targetRightEdge = totalWidth - margin.right; statusLegend.attr("transform", `translate(${targetRightEdge - lastRightEdge},${margin.top - 50})`); return wrap.node(); }2e:T3a01,// Taxonomy of Compute // Sourced from Adleman, Elowitz & Leibler, Gibson et al., White et al., Kagan et al., Jouppi et al., Merolla et al., Madsen et al., Arute et al., Monroe et al., Ebadi et al., Takeuchi et al., Likharev & Semenov, Pitowsky, Hogarth, Hamkins & Lewis. { const d3 = await require("d3@7"); const NB_BLUE = "#043463"; const BG = NB_BLUE; const GRID = "#ffffff"; const TEXT_ON_BLUE = "#ffffff"; const TEXT_SECONDARY_ON_BLUE = "rgba(255,255,255,0.72)"; const COMMERCIAL_GREEN = "#fe0100"; const RESEARCH_AMBER = "#ff8088"; const THEORY_GREY = "#ffffff"; const CELLAR_GREY = "#c9c9c9"; const CHIP_OPACITY = 1.0; const hardwareColumns = [ { id: "organic", label: "Organic", sub: "(Living Tissue)" }, { id: "charge", label: "Charge", sub: "(Silicon, PIM)" }, { id: "kinetic", label: "Kinetic", sub: "(Mechanical)" }, { id: "wave", label: "Wave", sub: "(EM, Acoustic, Magnetic, Plasma)" }, { id: "superconductor", label: "Superconductor", sub: "(Josephson)" }, { id: "atomic", label: "Atomic", sub: "(Trapped Ions, Neutral)" } ]; const softwareRows = [ { id: "fixed", label: "Fixed" }, { id: "universal", label: "Universal" }, { id: "learned", label: "Learned Weights" }, { id: "probabilistic", label: "Probabilistic" }, { id: "superposition", label: "Superposition" } ]; const cells = [ { hw: "organic", sw: "fixed", title: "Molecular Biology", examples: ["First RNA (~3.8 Gya)", "Adleman DNA (1994)", "Genetic circuits (2000)", "Engineered cells (2010)"], span: { sw: ["fixed", "universal"] }, status: "shipping" }, { hw: "organic", sw: "universal", title: "Brain Organoids", examples: ["Cortical Labs (2022)", "FinalSpark (2024)"], span: { hw: ["organic", "charge"] }, status: "prototype" }, { hw: "organic", sw: "learned", title: "Brains, Neurons", examples: ["Animal nervous systems", "(~600 Ma)"], status: "shipping" }, { hw: "charge", sw: "fixed", title: "Fixed-Function Silicon", examples: ["Calculator (1964)", "Bitcoin ASIC (2013)", "TPU (2016)"], status: "shipping" }, { hw: "charge", sw: "universal", title: "General-Purpose Silicon", examples: ["ENIAC (1945)", "GPU (1999)", "GPT-4 (2023)"], span: { sw: ["universal", "learned"] }, status: "shipping" }, { hw: "charge", sw: "learned", title: ["Memristive /", "Neuromorphic"], examples: ["TrueNorth (2014)", "Loihi 2 (2021)", "Mythic Analog (2022)"], status: "shipping" }, { hw: "charge", sw: "probabilistic", title: "Thermodynamic", examples: ["Normal Computing (2023)", "Extropic X0 (2024)"], status: "prototype" }, { hw: "kinetic", sw: "fixed", title: "Mechanical Calculator", examples: ["Antikythera (~100 BC)", "Babbage Diff. Engine (1822)"], status: "shipping" }, { hw: "kinetic", sw: "universal", title: "Mechanical Universal", examples: ["Analytical Engine (1837)"], status: "theoretical" }, { hw: "wave", sw: "fixed", title: "Photonic / Acoustic", examples: ["SAW Filters (1965)", "Lightmatter Envise (2021)"], status: "shipping" }, { hw: "wave", sw: "universal", title: "Optical CPU", examples: ["Caulfield Proposal (1989)"], status: "theoretical" }, { hw: "wave", sw: "learned", title: "Magnonic Networks", examples: ["Imec / Mainz (2014)"], status: "prototype" }, { hw: "wave", sw: "superposition", title: "Photonic Qubits", examples: ["Xanadu Borealis (2022)", "PsiQuantum Omega (2024)"], status: "prototype" }, { hw: "superconductor", sw: "universal", title: ["Reversible", "Superconducting"], examples: ["RSFQ (1985)", "AQFP (2014)"], status: "prototype" }, { hw: "superconductor", sw: "superposition", title: "Superconducting Qubits", examples: ["D-Wave One (2011)", "Google Willow (2024)", "IBM Nighthawk (2025)"], status: "prototype" }, { hw: "atomic", sw: "superposition", title: ["Trapped Ion /", "Neutral Atom"], examples: ["Wineland (1995)", "IonQ (2018)", "QuEra Aquila (2022)"], status: "prototype" } ]; const cellar = [ { title: "Carriers, not Computers", examples: ["Viruses", "USB sticks"] }, { title: "Reservoir-Style Physics", examples: ["Chemical", "Reservoir", "Slime mold", "Plasma"] }, { title: "Pre-Biotic / Cosmic", examples: ["Pre-biotic chemistry", "Cosmic computation"] } ]; const statusStyle = { shipping: { fill: COMMERCIAL_GREEN, label: "Commercial", textColor: "#ffffff" }, prototype: { fill: RESEARCH_AMBER, label: "Researching", textColor: "#0f0e0c" }, theoretical: { fill: THEORY_GREY, label: "Theoretical", textColor: "#0f0e0c" } }; // Layout const totalWidth = 1100; const PAD = 24; const softwareLabelGutter = 140; const bottomLabelGutter = 90; const matrixOriginX = PAD + softwareLabelGutter; const matrixOriginY = PAD; const matrixWidth = totalWidth - matrixOriginX - PAD; const colWidth = matrixWidth / hardwareColumns.length; const rowHeight = colWidth * 0.62; const matrixHeight = rowHeight * softwareRows.length; const legendBlockHeight = 36; const cellarBlockHeight = 130; const legendGap = 8; const cellarGap = 22; const totalHeight = matrixOriginY + matrixHeight + bottomLabelGutter + legendGap + legendBlockHeight + cellarGap + cellarBlockHeight + PAD; const cellCenterX = (colIdx) => matrixOriginX + colWidth / 2 + colIdx * colWidth; const cellCenterY = (rowIdx) => matrixOriginY + rowHeight / 2 + rowIdx * rowHeight; const svg = d3.create("svg") .attr("viewBox", [0, 0, totalWidth, totalHeight]) .attr("width", "100%") .style("background", BG) .style("font-family", "'Helvetica Neue', Arial, sans-serif"); const chipLayer = svg.append("g"); const labelLayer = svg.append("g"); const gridLayer = svg.append("g"); // Gridlines const GRID_STROKE_WIDTH = 1.4; const GRID_OPACITY = 0.85; for (let i = 0; i <= softwareRows.length; i++) { const y = matrixOriginY + i * rowHeight; gridLayer.append("line") .attr("x1", matrixOriginX).attr("y1", y) .attr("x2", matrixOriginX + matrixWidth).attr("y2", y) .attr("stroke", GRID).attr("stroke-width", GRID_STROKE_WIDTH).attr("opacity", GRID_OPACITY); } for (let i = 0; i <= hardwareColumns.length; i++) { const x = matrixOriginX + i * colWidth; gridLayer.append("line") .attr("x1", x).attr("y1", matrixOriginY) .attr("x2", x).attr("y2", matrixOriginY + matrixHeight) .attr("stroke", GRID).attr("stroke-width", GRID_STROKE_WIDTH).attr("opacity", GRID_OPACITY); } // Headers hardwareColumns.forEach((col, i) => { const x = cellCenterX(i); const y = matrixOriginY + matrixHeight + 24; labelLayer.append("text") .attr("x", x).attr("y", y) .attr("fill", TEXT_ON_BLUE) .attr("font-size", 13).attr("font-weight", 600) .attr("text-anchor", "middle") .text(col.label); labelLayer.append("text") .attr("x", x).attr("y", y + 16) .attr("fill", TEXT_SECONDARY_ON_BLUE) .attr("font-size", 11) .attr("text-anchor", "middle") .text(col.sub); }); labelLayer.append("text") .attr("x", matrixOriginX + matrixWidth / 2) .attr("y", matrixOriginY + matrixHeight + 76) .attr("fill", TEXT_ON_BLUE) .attr("font-size", 16).attr("font-weight", 700) .attr("text-anchor", "middle") .text("Hardware"); softwareRows.forEach((row, i) => { const x = matrixOriginX - 14; const y = cellCenterY(i) + 4; labelLayer.append("text") .attr("x", x).attr("y", y) .attr("fill", TEXT_ON_BLUE) .attr("font-size", 12).attr("font-weight", 600) .attr("text-anchor", "end") .text(row.label); }); const softwareLabelX = 32; labelLayer.append("text") .attr("x", softwareLabelX) .attr("y", matrixOriginY + matrixHeight / 2) .attr("fill", TEXT_ON_BLUE) .attr("font-size", 16).attr("font-weight", 700) .attr("text-anchor", "middle") .attr("transform", `rotate(-90, ${softwareLabelX}, ${matrixOriginY + matrixHeight / 2})`) .text("Software"); // Chip function drawChip(g, cx, cy, w, h, fill, opacity) { const r = Math.min(14, h / 2 - 2); g.append("rect") .attr("x", cx - w / 2) .attr("y", cy - h / 2) .attr("width", w) .attr("height", h) .attr("rx", r).attr("ry", r) .attr("fill", fill) .attr("fill-opacity", opacity) .attr("stroke", "rgba(0,0,0,0.18)") .attr("stroke-width", 0.8); } // Matrix cells.forEach(cell => { const colIdx = hardwareColumns.findIndex(h => h.id === cell.hw); const rowIdx = softwareRows.findIndex(r => r.id === cell.sw); let startCol = colIdx, endCol = colIdx; let startRow = rowIdx, endRow = rowIdx; if (cell.span && cell.span.hw) { startCol = hardwareColumns.findIndex(h => h.id === cell.span.hw[0]); endCol = hardwareColumns.findIndex(h => h.id === cell.span.hw[cell.span.hw.length - 1]); } if (cell.span && cell.span.sw) { startRow = softwareRows.findIndex(r => r.id === cell.span.sw[0]); endRow = softwareRows.findIndex(r => r.id === cell.span.sw[cell.span.sw.length - 1]); } const chipCx = (cellCenterX(startCol) + cellCenterX(endCol)) / 2; const chipCy = (cellCenterY(startRow) + cellCenterY(endRow)) / 2; const spanCols = endCol - startCol + 1; const spanRows = endRow - startRow + 1; const chipW = spanCols * colWidth - 14; const chipH = spanRows * rowHeight - 14; const style = statusStyle[cell.status]; const g = chipLayer.append("g").style("cursor", "pointer"); const titleLines = Array.isArray(cell.title) ? cell.title : [cell.title]; const titleFlat = titleLines.join(" "); g.append("title").text(titleFlat); drawChip(g, chipCx, chipCy, chipW, chipH, style.fill, CHIP_OPACITY); const textCx = cellCenterX(colIdx); const textCy = cellCenterY(rowIdx); const totalLines = titleLines.length + cell.examples.length; const titleYOffset = totalLines === 0 ? 4 : -((totalLines - 1) * 6) - 2; titleLines.forEach((line, ti) => { g.append("text") .attr("x", textCx).attr("y", textCy + titleYOffset + ti * 13) .attr("fill", style.textColor) .attr("font-size", 11.5).attr("font-weight", 700) .attr("text-anchor", "middle") .text(line); }); const examplesStartY = textCy + titleYOffset + (titleLines.length - 1) * 13 + 14; cell.examples.forEach((ex, i) => { g.append("text") .attr("x", textCx).attr("y", examplesStartY + i * 12) .attr("fill", style.textColor) .attr("font-size", 10) .attr("text-anchor", "middle") .text(ex); }); }); // Legend const legendY = matrixOriginY + matrixHeight + bottomLabelGutter + legendGap; const legendItems = [ { label: "Commercial", fill: COMMERCIAL_GREEN }, { label: "Researching", fill: RESEARCH_AMBER }, { label: "Theoretical", fill: THEORY_GREY } ]; const legendItemSpacing = 36; const legendChipW = 22; const legendChipH = 14; const legendTextOffset = 10; let legendTotalWidth = 0; const legendWidths = legendItems.map(item => { const textW = item.label.length * 7.2; const w = legendChipW + legendTextOffset + textW; legendTotalWidth += w; return w; }); legendTotalWidth += legendItemSpacing * (legendItems.length - 1); let cursorX = matrixOriginX + (matrixWidth - legendTotalWidth) / 2; const legendCenterY = legendY + legendBlockHeight / 2; legendItems.forEach((item, i) => { svg.append("rect") .attr("x", cursorX) .attr("y", legendCenterY - legendChipH / 2) .attr("width", legendChipW) .attr("height", legendChipH) .attr("rx", 5).attr("ry", 5) .attr("fill", item.fill) .attr("fill-opacity", CHIP_OPACITY) .attr("stroke", "rgba(0,0,0,0.18)") .attr("stroke-width", 0.8); svg.append("text") .attr("x", cursorX + legendChipW + legendTextOffset) .attr("y", legendCenterY + 4) .attr("fill", TEXT_ON_BLUE) .attr("font-size", 12) .attr("font-weight", 500) .text(item.label); cursorX += legendWidths[i] + legendItemSpacing; }); // Cellar const cellarTopY = legendY + legendBlockHeight + cellarGap; const dividerLeftX = PAD; const dividerRightX = totalWidth - PAD; svg.append("line") .attr("x1", dividerLeftX).attr("y1", cellarTopY - 8) .attr("x2", dividerRightX).attr("y2", cellarTopY - 8) .attr("stroke", GRID) .attr("stroke-width", 1.4) .attr("stroke-dasharray", "4 5") .attr("opacity", 0.7); const cellarHeaderX = PAD + 20; svg.append("text") .attr("x", cellarHeaderX).attr("y", cellarTopY + 18) .attr("fill", TEXT_ON_BLUE) .attr("font-size", 16).attr("font-weight", 700) .text("Not Computation"); svg.append("text") .attr("x", cellarHeaderX).attr("y", cellarTopY + 38) .attr("fill", TEXT_SECONDARY_ON_BLUE) .attr("font-size", 12) .text("Information storage, physics-as-computation, observer-imposed abstractions."); const cellarChipW = colWidth - 14; const cellarChipH = rowHeight - 14; const cellarChipsRowY = cellarTopY + 50 + cellarChipH / 2; const cellarSpanLeftX = matrixOriginX; const cellarSpanRightX = totalWidth - matrixOriginX; const cellarSpanWidth = cellarSpanRightX - cellarSpanLeftX; const cellarSlotWidth = cellarSpanWidth / cellar.length; cellar.forEach((item, i) => { const cx = cellarSpanLeftX + cellarSlotWidth * (i + 0.5); const cy = cellarChipsRowY; const r = Math.min(14, cellarChipH / 2 - 2); svg.append("rect") .attr("x", cx - cellarChipW / 2) .attr("y", cy - cellarChipH / 2) .attr("width", cellarChipW) .attr("height", cellarChipH) .attr("rx", r).attr("ry", r) .attr("fill", CELLAR_GREY) .attr("fill-opacity", CHIP_OPACITY) .attr("stroke", "rgba(0,0,0,0.18)") .attr("stroke-width", 0.8); const titleYOffset = -((item.examples.length - 1) * 6) - 2; svg.append("text") .attr("x", cx).attr("y", cy + titleYOffset) .attr("fill", "#0f0e0c") .attr("font-size", 11.5).attr("font-weight", 700) .attr("text-anchor", "middle") .text(item.title); item.examples.forEach((ex, j) => { svg.append("text") .attr("x", cx).attr("y", cy + titleYOffset + 14 + j * 12) .attr("fill", "#0f0e0c") .attr("font-size", 10) .attr("text-anchor", "middle") .text(ex); }); }); return svg.node(); }2f:[] 2b:{"id":"d8cbbc12efa06726","slug":null,"title":"Efficient Frontier of Compute","thumbnail":"3275580bd932dcd19ae4575d0a24399b8d17748aa9840456daabf96bf6605a5a","access_level":"public","create_time":"$D2026-07-15T21:58:20.164Z","update_time":"$D2026-08-16T16:35:32.117Z","view_time":null,"first_public_time":"$D2026-08-02T02:32:49.717Z","has_unpublished_edits":null,"fork":false,"stars":1,"starred":false,"owner":{"type":"team","id":"76a2df239d009536","login":"northbranchmedia","name":"North Branch Media","avatar_url":"https://avatars-next.observableusercontent.com/76819e141d8360b0fb326704433556feaf213b308230b2c1906d59298189200a","tier":"starter_2024"},"version":119,"publish_version":null,"role":null,"authored":false,"authors":[{"type":"individual","id":"1a4a5420f054ddf6","login":"north-branch-media","name":"North Branch Media","avatar_url":"https://avatars-next.observableusercontent.com/c80485c224f42cb4baa7e0bcefcda23343f268e832e216af1cc503d13623ea1c","tier":null}],"members":[{"type":"team","id":"76a2df239d009536","login":"northbranchmedia","name":"North Branch Media","avatar_url":"https://avatars-next.observableusercontent.com/76819e141d8360b0fb326704433556feaf213b308230b2c1906d59298189200a","tier":"starter_2024","role":"editor"},{"type":"individual","id":"1a4a5420f054ddf6","login":"north-branch-media","name":"North Branch Media","avatar_url":"https://avatars-next.observableusercontent.com/c80485c224f42cb4baa7e0bcefcda23343f268e832e216af1cc503d13623ea1c","tier":null,"role":"editor"}],"comments":[],"custom_thumbnail":false,"trashed":false,"lapsed":null,"subscribed":null,"forks":0,"collection":{"id":"39e969d39e7db6ae","slug":"2026","title":"2026","owner":{"id":"76a2df239d009536","login":"northbranchmedia","name":"North Branch Media"}},"type":"notebook","body":{"title":"Efficient Frontier of Compute","cells":[{"id":0,"value":"# Efficient Frontier of Compute","pinned":false,"hidden":false,"mode":"md","author":"$undefined","summary":"$undefined","output":"","database":"$undefined"},{"id":4,"value":"$2d","pinned":true,"hidden":false,"mode":"ojs","author":"$undefined","summary":"$undefined","output":"$undefined","database":"$undefined"},{"id":15,"value":"$2e","pinned":true,"hidden":false,"mode":"ojs","author":"$undefined","summary":"$undefined","output":"$undefined","database":"$undefined"}],"files":"$W2f","stdlib":"1","titling":"auto"}} 2c:"$2b:body"