{
  "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": 63,
  "pageCount": 313,
  "prevPage": 62,
  "nextPage": 64,
  "slideType": "industry_trends",
  "function": "analyze_data",
  "density": "overcrowded",
  "nDataPoints": 12,
  "notes": "The slide compares Transformer++, Mamba, Hyena, and StripedHyena architectures in the context of genomic sequence modeling.",
  "elementsJson": [
    "headline_text",
    "bullet_list",
    "line_chart",
    "paragraph"
  ],
  "metadataConfidence": 1,
  "imagePath": null,
  "slideHref": "/slides/019dd923-5e88-73ef-bd5d-0d0f98caffe1/63",
  "deckHref": "/decks/019dd923-5e88-73ef-bd5d-0d0f98caffe1",
  "deckJsonHref": "/decks/019dd923-5e88-73ef-bd5d-0d0f98caffe1.json",
  "deckAnchorHref": "/decks/019dd923-5e88-73ef-bd5d-0d0f98caffe1#slide-63",
  "components": [
    {
      "bbox": null,
      "kind": "callout",
      "text": "Next-token modeling learns real biological dependencies from DNA, and Evo's scaling study shows smooth, compute-predictable loss improvements with more data, parameters, and context, plus clear architecture effects.",
      "attrs": null,
      "subkind": null,
      "toolName": "Visual emphasis",
      "toolSlug": "visual-emphasis",
      "confidence": null,
      "componentId": "019dd952-41ac-7180-a821-02a43e45510b",
      "frameworkName": null,
      "frameworkSlug": null
    },
    {
      "bbox": {
        "h": 0.25,
        "w": 0.15,
        "x": 0.3,
        "y": 0.62
      },
      "kind": "chart",
      "text": "Architecture comparison charts (Transformer++, Mamba, Hyena, StripedHyena)",
      "attrs": null,
      "subkind": "line",
      "toolName": null,
      "toolSlug": null,
      "confidence": null,
      "componentId": "91169f3f-9d7b-405a-8788-e38a34c3890d",
      "frameworkName": null,
      "frameworkSlug": null
    },
    {
      "bbox": {
        "h": 0.3,
        "w": 0.18,
        "x": 0.78,
        "y": 0.32
      },
      "kind": "chart",
      "text": "Evo-2 validation perplexity chart",
      "attrs": null,
      "subkind": "line",
      "toolName": null,
      "toolSlug": null,
      "confidence": null,
      "componentId": "91cf3519-028a-44e9-ac6b-8f072a9079d9",
      "frameworkName": null,
      "frameworkSlug": null
    },
    {
      "bbox": {
        "h": 0.25,
        "w": 0.25,
        "x": 0.05,
        "y": 0.62
      },
      "kind": "chart",
      "text": "Compute-optimal scaling chart",
      "attrs": null,
      "subkind": "line",
      "toolName": null,
      "toolSlug": null,
      "confidence": null,
      "componentId": "ea01fc51-bc7a-47c7-9a45-3c8aa1461d08",
      "frameworkName": null,
      "frameworkSlug": null
    },
    {
      "bbox": {
        "h": 0.25,
        "w": 0.25,
        "x": 0.5,
        "y": 0.32
      },
      "kind": "list",
      "text": "Evo-2 (Feb '25) pushes further by training 40B and 7B models on 9.3T and 2.4T tokens, respectively, and extending context to 1M. Validation perplexity improves with both model size and context, and long-context recall remains effective at 1M.",
      "attrs": null,
      "subkind": "bullet",
      "toolName": null,
      "toolSlug": null,
      "confidence": null,
      "componentId": "8ad9a3ce-5cb1-461a-b2e7-969e51461ad7",
      "frameworkName": null,
      "frameworkSlug": null
    },
    {
      "bbox": {
        "h": 0.25,
        "w": 0.45,
        "x": 0.02,
        "y": 0.32
      },
      "kind": "list",
      "text": "Evo (Nov '24) is trained on ~300B nucleotides, byte-tokenized, with context up to 131k. Along the compute-optimal frontier, the Hyena family (input-dependent long convolutions with a few attention layers) delivers lower PPLX/FLOP and more stable training than Transformer++/Mamba",
      "attrs": null,
      "subkind": "bullet",
      "toolName": null,
      "toolSlug": null,
      "confidence": null,
      "componentId": "e0c61dbe-eafb-4da9-93c0-7c0e1d5af257",
      "frameworkName": null,
      "frameworkSlug": null
    },
    {
      "bbox": null,
      "kind": "metric",
      "text": "Validation perplexity",
      "attrs": null,
      "subkind": "primary",
      "toolName": "Quantification",
      "toolSlug": "quantification",
      "confidence": null,
      "componentId": "019dd952-41ac-7180-a821-06d0f326f076",
      "frameworkName": null,
      "frameworkSlug": null
    },
    {
      "bbox": {
        "h": 0.08,
        "w": 0.95,
        "x": 0.02,
        "y": 0.22
      },
      "kind": "paragraph",
      "text": "Next-token modeling learns real biological dependencies from DNA, and Evo's scaling study shows smooth, compute-predictable loss improvements with more data, parameters, and context, plus clear architecture effects.",
      "attrs": null,
      "subkind": "paragraph",
      "toolName": null,
      "toolSlug": null,
      "confidence": null,
      "componentId": "629b07b5-f314-4f88-933c-e5543c457c5c",
      "frameworkName": null,
      "frameworkSlug": null
    },
    {
      "bbox": {
        "h": 0.05,
        "w": 0.7,
        "x": 0.02,
        "y": 0.14
      },
      "kind": "title",
      "text": "Scaling laws in genomics: predictable gains with compute, data, and context",
      "attrs": null,
      "subkind": "headline",
      "toolName": null,
      "toolSlug": null,
      "confidence": null,
      "componentId": "d9b9eb2b-f4be-4241-957d-9a666fe43f79",
      "frameworkName": null,
      "frameworkSlug": null
    }
  ],
  "metrics": [],
  "tools": [
    {
      "name": "Chart Selection Guide",
      "slug": "chart-selection-guide",
      "agent": "Designer",
      "layer": "slide",
      "matchId": "07424641-6451-46a4-93a6-4b85e61125dd",
      "evidence": "The slide includes various charts (line, scatter) to illustrate the relationship between architecture, compute, and performance.",
      "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
}