{
  "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": 62,
  "pageCount": 313,
  "prevPage": 61,
  "nextPage": 63,
  "slideType": "case_study",
  "function": "illustrate_case",
  "density": "dense",
  "nDataPoints": 4,
  "notes": "The slide describes a system that automates scientific method invention by treating it as a search problem.",
  "elementsJson": [
    "headline_text",
    "action_title",
    "paragraph",
    "bullet_list",
    "process_diagram",
    "bar_chart_vertical"
  ],
  "metadataConfidence": 0.95,
  "imagePath": null,
  "slideHref": "/slides/019dd923-5e88-73ef-bd5d-0d0f98caffe1/62",
  "deckHref": "/decks/019dd923-5e88-73ef-bd5d-0d0f98caffe1",
  "deckJsonHref": "/decks/019dd923-5e88-73ef-bd5d-0d0f98caffe1.json",
  "deckAnchorHref": "/decks/019dd923-5e88-73ef-bd5d-0d0f98caffe1#slide-62",
  "components": [
    {
      "bbox": null,
      "kind": "callout",
      "text": "An LLM guided by a modified tree-search (“code mutation system”) generates, runs, and iterates code until it beats established leaderboards.",
      "attrs": null,
      "subkind": null,
      "toolName": "Visual emphasis",
      "toolSlug": "visual-emphasis",
      "confidence": null,
      "componentId": "019dd952-41ac-7180-a821-784469f6d331",
      "frameworkName": null,
      "frameworkSlug": null
    },
    {
      "bbox": {
        "h": 0.25,
        "w": 0.25,
        "x": 0.65,
        "y": 0.65
      },
      "kind": "chart",
      "text": "Public leaderboard percentile performance",
      "attrs": null,
      "subkind": "bar-vertical",
      "toolName": null,
      "toolSlug": null,
      "confidence": null,
      "componentId": "cec00933-c485-42eb-8e74-273baf757b6f",
      "frameworkName": null,
      "frameworkSlug": null
    },
    {
      "bbox": {
        "h": 0.4,
        "w": 0.3,
        "x": 0.65,
        "y": 0.45
      },
      "kind": "diagram",
      "text": "Tree search and system architecture flow",
      "attrs": null,
      "subkind": "process",
      "toolName": null,
      "toolSlug": null,
      "confidence": null,
      "componentId": "ea35044a-3f26-4cb5-8bd2-92d695b6108a",
      "frameworkName": null,
      "frameworkSlug": null
    },
    {
      "bbox": {
        "h": 0.45,
        "w": 0.6,
        "x": 0.05,
        "y": 0.4
      },
      "kind": "list",
      "text": "On the OpenProblems single cell RNA-seq integration benchmark, 40 of 87 generated methods including recombinations and ideas seeded by “Deep Research” and “AI co-scientist” outperform all published leaderboard entries.\nFor numerical integration, the evolved algorithm succeeds on 17/19 hard integrals (≤3% error) where scipy.integrate.quad() fails on all 19, via adaptive domain partitioning plus Euler series acceleration.\nThe system competes on the CDC COVID-19 Forecast Hub and reproduces/innovates over strong AR, GBM, and mechanistic models. it also tackles remote-sensing segmentation (DLRSD) and other tasks, showing breadth beyond a single field.",
      "attrs": null,
      "subkind": "bullet",
      "toolName": null,
      "toolSlug": null,
      "confidence": null,
      "componentId": "4ebce646-76c7-4194-80aa-29761ad0a614",
      "frameworkName": null,
      "frameworkSlug": null
    },
    {
      "bbox": null,
      "kind": "metric",
      "text": "Public leaderboard percentile: 40/87",
      "attrs": null,
      "subkind": "primary",
      "toolName": "Quantification",
      "toolSlug": "quantification",
      "confidence": null,
      "componentId": "019dd952-41ac-7180-a821-7cf812abdd73",
      "frameworkName": null,
      "frameworkSlug": null
    },
    {
      "bbox": {
        "h": 0.15,
        "w": 0.9,
        "x": 0.05,
        "y": 0.22
      },
      "kind": "paragraph",
      "text": "An LLM guided by a modified tree-search (“code mutation system”) generates, runs, and iterates code until it beats established leaderboards. The system recombines ideas, proposes new ones, and rigorously scores each attempt on public benchmarks, turning method invention into an automated search problem. Results span single-cell RNA-seq integration, COVID-19 forecasting, remote sensing, and numerical analysis.",
      "attrs": null,
      "subkind": "paragraph",
      "toolName": null,
      "toolSlug": null,
      "confidence": null,
      "componentId": "7154ab0b-b30f-4fb0-8f3e-6f9497b8df9e",
      "frameworkName": null,
      "frameworkSlug": null
    },
    {
      "bbox": {
        "h": 0.05,
        "w": 0.8,
        "x": 0.05,
        "y": 0.14
      },
      "kind": "title",
      "text": "LLM-driven tree search writes expert-level scientific software across domains",
      "attrs": null,
      "subkind": "headline",
      "toolName": null,
      "toolSlug": null,
      "confidence": null,
      "componentId": "0bee5f4d-6be3-4327-a1a3-3359c4dfc827",
      "frameworkName": null,
      "frameworkSlug": null
    }
  ],
  "metrics": [],
  "tools": [
    {
      "name": "Hypothesis-Driven Structure",
      "slug": "hypothesis-driven-structure",
      "agent": "Architect",
      "layer": "slide",
      "matchId": "e952cc2a-085e-49bd-ab87-dde6d14555e3",
      "evidence": "The slide mentions a modified tree-search (“ code mutation system”) generating, running, and iterating code until it beats established leaderboards.",
      "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
}