{
  "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": 23,
  "pageCount": 313,
  "prevPage": 22,
  "nextPage": 24,
  "slideType": "key_takeaways",
  "function": "summarize",
  "density": "dense",
  "nDataPoints": 3,
  "notes": "The slide highlights three specific research papers/projects: GSM-Symbolic, DataAlchemy, and XReasoning.",
  "elementsJson": [
    "headline_text",
    "paragraph",
    "bullet_list"
  ],
  "metadataConfidence": 1,
  "imagePath": null,
  "slideHref": "/slides/019dd923-5e88-73ef-bd5d-0d0f98caffe1/23",
  "deckHref": "/decks/019dd923-5e88-73ef-bd5d-0d0f98caffe1",
  "deckJsonHref": "/decks/019dd923-5e88-73ef-bd5d-0d0f98caffe1.json",
  "deckAnchorHref": "/decks/019dd923-5e88-73ef-bd5d-0d0f98caffe1#slide-23",
  "components": [
    {
      "bbox": null,
      "kind": "callout",
      "text": "This suggests that models do template-matching rather than true algebraic reasoning.",
      "attrs": null,
      "subkind": null,
      "toolName": "Visual emphasis",
      "toolSlug": "visual-emphasis",
      "confidence": null,
      "componentId": "019dd952-41ac-7180-a81b-0d32ee2517f3",
      "frameworkName": null,
      "frameworkSlug": null
    },
    {
      "bbox": {
        "h": 0.4,
        "w": 0.9,
        "x": 0.05,
        "y": 0.4
      },
      "kind": "list",
      "text": "Apple’s GSM-Symbolic work shows that accuracy drops sharply when only the numeric instantiation changes, and adding a single seemingly relevant clause can cut performance by ~65%. This suggests that models do template-matching rather than true algebraic reasoning.\nWork from ASU’s DataAlchemy found that the CoT helps in-distribution but collapses when test tasks, chain length, or CoT format deviate from training. Longer, well-worded traces often mask incorrect logic.\nFinally, Groningen/Harvard/MGH/Amsterdam’s XReasoning shows how prompt-forcing the model to reason in the user’s language lifts match rates to ~98% on hard sets but reduces accuracy by 9–13 points. Using 100–250 example post-training improves language match but the accuracy penalty remains.",
      "attrs": null,
      "subkind": "bullet",
      "toolName": null,
      "toolSlug": null,
      "confidence": null,
      "componentId": "d2ae3b4f-a46c-4aa5-ac55-645c47cb3894",
      "frameworkName": null,
      "frameworkSlug": null
    },
    {
      "bbox": null,
      "kind": "metric",
      "text": "accuracy: 65%",
      "attrs": null,
      "subkind": "primary",
      "toolName": "Quantification",
      "toolSlug": "quantification",
      "confidence": null,
      "componentId": "019dd952-41ac-7180-a81b-13f2f1895e6a",
      "frameworkName": null,
      "frameworkSlug": null
    },
    {
      "bbox": {
        "h": 0.15,
        "w": 0.9,
        "x": 0.05,
        "y": 0.22
      },
      "kind": "paragraph",
      "text": "Reasoning also degrades under mild distribution changes. Changing numbers or adding one innocuous clause slashes math accuracy, while shifting the length/format of chains-of-thought makes models produce fluent but incoherent steps. Forcing the model to “think” in a user’s language raises readability but lowers accuracy. These effects persist at larger scales and after light post-training.",
      "attrs": null,
      "subkind": "paragraph",
      "toolName": null,
      "toolSlug": null,
      "confidence": null,
      "componentId": "beefd033-71e9-4f7a-9ef1-ac712159a039",
      "frameworkName": null,
      "frameworkSlug": null
    },
    {
      "bbox": {
        "h": 0.05,
        "w": 0.5,
        "x": 0.05,
        "y": 0.14
      },
      "kind": "title",
      "text": "How reasoning breaks: small shifts cause big failures",
      "attrs": null,
      "subkind": "headline",
      "toolName": null,
      "toolSlug": null,
      "confidence": null,
      "componentId": "f821fc90-40ab-490d-84b3-2e97557bccc1",
      "frameworkName": null,
      "frameworkSlug": null
    }
  ],
  "metrics": [],
  "tools": [
    {
      "name": "Concrete Language",
      "slug": "concrete-language",
      "agent": "Storyteller",
      "layer": "slide",
      "matchId": "019dd95a-1055-74e0-a95b-f993f1e17c69",
      "evidence": "'65%' performance drop concretizes 'small shifts cause big failures'.",
      "confidence": 85
    }
  ],
  "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": 30,
      "from": 13,
      "name": "Zoom In",
      "slug": "06-zoom-in",
      "bestFor": "Technical deep-dives, case studies, detailed analysis",
      "matchId": "019dd95a-07fe-70ce-8d3e-9252f6f92bd7",
      "evidence": "Big picture 'reasoning race' (p.13) zooms into o1, R1, DeepSeek, parallel reasoning, CoT mechanics.",
      "position": 2,
      "objective": "Drill from reasoning paradigm into specific reasoning research findings",
      "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"
    },
    {
      "to": 23,
      "from": 20,
      "name": "Myth Buster",
      "slug": "12-myth-buster",
      "bestFor": "Rebranding, changing market perception, correcting false assumptions",
      "matchId": "019dd95a-07fe-70ce-8d3e-94ced6245fc4",
      "evidence": "'The illusion of reasoning gains' (p.20), 'How reasoning breaks' (pp.22-23) directly challenge consensus.",
      "position": 3,
      "objective": "Bust the 'reasoning gains are real' belief",
      "structure": "The Common Belief -> The Friction/Failure of that Belief -> The New Truth",
      "confidence": 80,
      "description": "Address a common misconception head-on to clear the room for a new truth"
    }
  ],
  "imagePathAlt": null,
  "thumbSrc": null,
  "thumbSrcAlt": null,
  "locked": true
}