{
  "docId": "019dd923-5e88-73ef-bd5d-0d0f98caffe1",
  "docSlug": "5df6bd1b0447b5f6",
  "documentTitle": "2025 Air Street Capital The State of AI Report 2025",
  "authorId": "AirStreetCapital",
  "authorName": "Air Street Capital",
  "documentKindSlug": "consulting-deck",
  "documentKindLabel": "Consulting deck",
  "sourceTypeSlug": "vc_research",
  "sourceTypeLabel": "VC research",
  "presentationDate": null,
  "orientation": "landscape",
  "aspectRatio": 1.777,
  "pageNumber": 64,
  "pageCount": 313,
  "prevPage": 63,
  "nextPage": 65,
  "slideType": "industry_trends",
  "function": "analyze_data",
  "density": "dense",
  "nDataPoints": 3,
  "notes": "The slide uses three distinct charts to demonstrate scaling laws, sequence space generation, and alignment benefits for protein LMs.",
  "elementsJson": [
    "headline_text",
    "paragraph",
    "line_chart",
    "heatmap",
    "line_chart"
  ],
  "metadataConfidence": 1,
  "imagePath": null,
  "slideHref": "/slides/019dd923-5e88-73ef-bd5d-0d0f98caffe1/64",
  "deckHref": "/decks/019dd923-5e88-73ef-bd5d-0d0f98caffe1",
  "deckJsonHref": "/decks/019dd923-5e88-73ef-bd5d-0d0f98caffe1.json",
  "deckAnchorHref": "/decks/019dd923-5e88-73ef-bd5d-0d0f98caffe1#slide-64",
  "components": [
    {
      "bbox": null,
      "kind": "callout",
      "text": "Protein LMs also obey smooth scaling laws.",
      "attrs": null,
      "subkind": null,
      "toolName": "Visual emphasis",
      "toolSlug": "visual-emphasis",
      "confidence": null,
      "componentId": "019dd952-41ac-7180-a820-de44b9b91d27",
      "frameworkName": null,
      "frameworkSlug": null
    },
    {
      "bbox": {
        "h": 0.45,
        "w": 0.28,
        "x": 0.68,
        "y": 0.45
      },
      "kind": "chart",
      "text": "Fitness Prediction Spearman vs Model Scale",
      "attrs": null,
      "subkind": "line",
      "toolName": null,
      "toolSlug": null,
      "confidence": null,
      "componentId": "8323f883-f75f-44d8-84cc-72a42488ab38",
      "frameworkName": null,
      "frameworkSlug": null
    },
    {
      "bbox": {
        "h": 0.45,
        "w": 0.3,
        "x": 0.05,
        "y": 0.45
      },
      "kind": "chart",
      "text": "Validation Loss vs Training Compute",
      "attrs": null,
      "subkind": "line",
      "toolName": null,
      "toolSlug": null,
      "confidence": null,
      "componentId": "fa35214c-2dea-4edd-9e21-3a09c4d92807",
      "frameworkName": null,
      "frameworkSlug": null
    },
    {
      "bbox": {
        "h": 0.45,
        "w": 0.25,
        "x": 0.38,
        "y": 0.45
      },
      "kind": "chart",
      "text": "Experimental expression sequence space",
      "attrs": null,
      "subkind": "scatter",
      "toolName": null,
      "toolSlug": null,
      "confidence": null,
      "componentId": "fb769de2-86a3-4a54-909c-4a2a1fae4232",
      "frameworkName": null,
      "frameworkSlug": null
    },
    {
      "bbox": null,
      "kind": "metric",
      "text": "Validation Loss: 46B",
      "attrs": null,
      "subkind": "primary",
      "toolName": "Quantification",
      "toolSlug": "quantification",
      "confidence": null,
      "componentId": "019dd952-41ac-7180-a820-e1e2a231e4b3",
      "frameworkName": null,
      "frameworkSlug": null
    },
    {
      "bbox": {
        "h": 0.12,
        "w": 0.95,
        "x": 0.03,
        "y": 0.21
      },
      "kind": "paragraph",
      "text": "Protein LMs also obey smooth scaling laws. Profluent's ProGen3 derives a compute-optimal frontier for sparse autoregressive PLMs and then scales to a 46B MoE trained on PPA-1: 3.4B full-length proteins (1.1T tokens), ultimately training on 1.5T tokens (left chart). Larger models generate viable proteins across broader sequence space (middle chart), and alignment lifts performance most at larger scale (right chart).",
      "attrs": null,
      "subkind": "paragraph",
      "toolName": null,
      "toolSlug": null,
      "confidence": null,
      "componentId": "0431d075-555a-4012-960d-f04c907c15db",
      "frameworkName": null,
      "frameworkSlug": null
    },
    {
      "bbox": {
        "h": 0.04,
        "w": 0.65,
        "x": 0.03,
        "y": 0.14
      },
      "kind": "title",
      "text": "Scaling laws for proteins unlock broader and more useful generation",
      "attrs": null,
      "subkind": "headline",
      "toolName": null,
      "toolSlug": null,
      "confidence": null,
      "componentId": "b4ad4661-39b8-44c9-9e79-919e1ac686cb",
      "frameworkName": null,
      "frameworkSlug": null
    }
  ],
  "metrics": [],
  "tools": [
    {
      "name": "Waterfall chart",
      "slug": "waterfall-chart",
      "agent": null,
      "layer": "slide",
      "matchId": "293dd6fc-2774-4350-9deb-798c24ce9c9c",
      "evidence": "The slide mentions a compute-optimal frontier for sparse autoregressive PLMs and a waterfall-like presentation of results.",
      "confidence": 0.7
    }
  ],
  "frameworks": [],
  "arcBeats": [
    {
      "to": 189,
      "from": 12,
      "beatId": "019dd95a-0682-776c-8e35-9f380673831f",
      "arcName": "The Triple Take",
      "arcSlug": "triple-take",
      "beatName": "The Facts (What)",
      "beatSlug": "triple-take-the-facts-what",
      "evidence": "Sections 1-2 lay out research findings and industry data with charts and case studies.",
      "position": 1,
      "confidence": 70,
      "parentBeatName": "Setup",
      "parentBeatSlug": "setup"
    },
    {
      "to": 89,
      "from": 12,
      "beatId": "019dd95a-0682-776c-8e35-ac68557d958b",
      "arcName": "The Onion",
      "arcSlug": "onion",
      "beatName": "First Layer",
      "beatSlug": "onion-first-layer",
      "evidence": "Research section unpacks technical layer beneath the headlines.",
      "position": 2,
      "confidence": 45,
      "parentBeatName": "Development",
      "parentBeatSlug": "development"
    }
  ],
  "loops": [
    {
      "to": 67,
      "from": 55,
      "name": "Zoom In",
      "slug": "06-zoom-in",
      "bestFor": "Technical deep-dives, case studies, detailed analysis",
      "matchId": "019dd95a-07fe-70ce-8d3e-9d32ca243ef1",
      "evidence": "Sequence of case studies: AlphaEvolve, UMA, MatterGen, AMIE, AlphaFold reproductions all illustrate AI-as-scientist.",
      "position": 5,
      "objective": "Survey AI's rise as scientific collaborator across biology/chemistry/medicine",
      "structure": "The Big Picture -> Key Area of Focus -> Specific Detail -> Implication",
      "confidence": 75,
      "description": "Start broad, then progressively focus on specific details that prove your point"
    }
  ],
  "imagePathAlt": null,
  "thumbSrc": null,
  "thumbSrcAlt": null,
  "locked": true
}