{
  "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": 60,
  "pageCount": 313,
  "prevPage": 59,
  "nextPage": 61,
  "slideType": "industry_trends",
  "function": "summarize",
  "density": "dense",
  "nDataPoints": 8,
  "notes": "The chart shows the evolution of chemical synthesis benchmark scores over time for various LLM providers.",
  "elementsJson": [
    "bullet_list",
    "paragraph",
    "scatter_plot"
  ],
  "metadataConfidence": 1,
  "imagePath": null,
  "slideHref": "/slides/019dd923-5e88-73ef-bd5d-0d0f98caffe1/60",
  "deckHref": "/decks/019dd923-5e88-73ef-bd5d-0d0f98caffe1",
  "deckJsonHref": "/decks/019dd923-5e88-73ef-bd5d-0d0f98caffe1.json",
  "deckAnchorHref": "/decks/019dd923-5e88-73ef-bd5d-0d0f98caffe1#slide-60",
  "components": [
    {
      "bbox": null,
      "kind": "callout",
      "text": "Chemical modeling has shifted from task-specific predictors to general LLMs that reason about synthesis strategy and mechanisms.",
      "attrs": null,
      "subkind": null,
      "toolName": "Visual emphasis",
      "toolSlug": "visual-emphasis",
      "confidence": null,
      "componentId": "019dd952-41ac-7180-a820-bc7ed0660f0f",
      "frameworkName": null,
      "frameworkSlug": null
    },
    {
      "bbox": {
        "h": 0.5,
        "w": 0.22,
        "x": 0.75,
        "y": 0.4
      },
      "kind": "chart",
      "text": "Chemical synthesis benchmark score over time",
      "attrs": null,
      "subkind": "scatter",
      "toolName": null,
      "toolSlug": null,
      "confidence": null,
      "componentId": "2c8c8209-93d8-4cf3-bbec-6e52a75cf89b",
      "frameworkName": null,
      "frameworkSlug": null
    },
    {
      "bbox": {
        "h": 0.45,
        "w": 0.7,
        "x": 0.02,
        "y": 0.42
      },
      "kind": "list",
      "text": "ChemBench finds frontier models (e.g., o1-preview; open Llama-3.1-405B close behind) outperform the best human chemists on aggregate, with performance rising with model size. Retrieval alone doesn’t fix knowledge-heavy failures.\nThe current best approach uses a large LLMs as a strategic evaluator plugged into search (incl. MCTS): it judges routes and mechanisms from natural-language constraints. Newer, larger models (e.g., Gemini-2.5-Pro) lead; strong open options (e.g., DeepSeek-r1) are close.\nThis LLM-as-judge + search pattern brings human-like planning (protecting-group timing, ring-formation order) without forcing the LLM to emit SMILES (still challenging) and it scales as LLMs improve, and as inference time increases.",
      "attrs": null,
      "subkind": "bullet",
      "toolName": null,
      "toolSlug": null,
      "confidence": null,
      "componentId": "5ac1a830-d5ea-4a44-8b3e-e32a3c0a1b8e",
      "frameworkName": null,
      "frameworkSlug": null
    },
    {
      "bbox": null,
      "kind": "metric",
      "text": "Chemical synthesis benchmark score",
      "attrs": null,
      "subkind": "primary",
      "toolName": "Quantification",
      "toolSlug": "quantification",
      "confidence": null,
      "componentId": "019dd952-41ac-7180-a820-c32c58ddf518",
      "frameworkName": null,
      "frameworkSlug": null
    },
    {
      "bbox": {
        "h": 0.18,
        "w": 0.7,
        "x": 0.02,
        "y": 0.21
      },
      "kind": "paragraph",
      "text": "Chemical modeling has shifted from task-specific predictors to general LLMs that reason about synthesis strategy and mechanisms. The strongest results now come from large, general-purpose transformers used as “reasoning engines” and paired with classical search, rather than chemistry-specific generators. Benchmarks also show these models matching or beating expert chemists on curated Q&A while still missing knowledge-intensive and multi-modal edge cases.",
      "attrs": null,
      "subkind": "paragraph",
      "toolName": null,
      "toolSlug": null,
      "confidence": null,
      "componentId": "8354b11d-6db2-425a-97e6-489b88fe2624",
      "frameworkName": null,
      "frameworkSlug": null
    },
    {
      "bbox": {
        "h": 0.05,
        "w": 0.7,
        "x": 0.02,
        "y": 0.14
      },
      "kind": "title",
      "text": "Language models in Chemistry: from property predictors to strategy-aware planners",
      "attrs": null,
      "subkind": "headline",
      "toolName": null,
      "toolSlug": null,
      "confidence": null,
      "componentId": "fc176f80-1328-4f4b-8e8f-ab87e5641337",
      "frameworkName": null,
      "frameworkSlug": null
    }
  ],
  "metrics": [],
  "tools": [
    {
      "name": "Audience Definition",
      "slug": "audience-definition",
      "agent": "Storyteller",
      "layer": "slide",
      "matchId": "98b052cc-726c-4204-be79-7c0f0b742621",
      "evidence": "The slide is about language models in Chemistry, indicating an audience interested in chemistry and AI.",
      "confidence": 0.7
    },
    {
      "name": "Data Story Arc",
      "slug": "data-story-arc",
      "agent": "Storyteller",
      "layer": "slide",
      "matchId": "19a4ef9e-f859-4862-91e2-978946a15583",
      "evidence": "The slide presents a story about the evolution of language models in chemistry, using data to support the narrative.",
      "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
}