{
  "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": 71,
  "pageCount": 313,
  "prevPage": 70,
  "nextPage": 72,
  "slideType": "problem_statement",
  "function": "diagnose",
  "density": "dense",
  "nDataPoints": 6,
  "notes": "The slide highlights how LMArena data is being manipulated and how proprietary models have an unfair advantage in benchmark performance.",
  "elementsJson": [
    "headline_text",
    "bullet_list",
    "bubble_chart"
  ],
  "metadataConfidence": 1,
  "imagePath": null,
  "slideHref": "/slides/019dd923-5e88-73ef-bd5d-0d0f98caffe1/71",
  "deckHref": "/decks/019dd923-5e88-73ef-bd5d-0d0f98caffe1",
  "deckJsonHref": "/decks/019dd923-5e88-73ef-bd5d-0d0f98caffe1.json",
  "deckAnchorHref": "/decks/019dd923-5e88-73ef-bd5d-0d0f98caffe1#slide-71",
  "components": [
    {
      "bbox": null,
      "kind": "callout",
      "text": "Researchers revealed systematic manipulation of LMArena as Meta tested 27 private Llama-4 variants before cherry-picking the winner: testing 10 model variants yields a 100-point boost.",
      "attrs": null,
      "subkind": null,
      "toolName": "Visual emphasis",
      "toolSlug": "visual-emphasis",
      "confidence": null,
      "componentId": "019dd952-41ac-7180-a821-9fd8c7f37c17",
      "frameworkName": null,
      "frameworkSlug": null
    },
    {
      "bbox": {
        "h": 0.45,
        "w": 0.45,
        "x": 0.52,
        "y": 0.3
      },
      "kind": "chart",
      "text": "Bubble chart showing Number of Models vs Maximum Arena Score, categorized by model type (Open Source, Open Weights, Proprietary) and size of bubbles representing Approx Battles.",
      "attrs": null,
      "subkind": "bubble",
      "toolName": null,
      "toolSlug": null,
      "confidence": null,
      "componentId": "a5e8c656-d04c-4703-8ac0-1fbf617fa10f",
      "frameworkName": null,
      "frameworkSlug": null
    },
    {
      "bbox": {
        "h": 0.55,
        "w": 0.48,
        "x": 0.02,
        "y": 0.3
      },
      "kind": "list",
      "text": "OpenAI and Google are hoovering up 40% of all Arena data while 83 open models fight over 30% of the scraps.\nBig Tech gets 68 times more data access than academic labs.\nTraining on Arena data doubles your win rate.\nContamination audits needed.",
      "attrs": null,
      "subkind": "bullet",
      "toolName": null,
      "toolSlug": null,
      "confidence": null,
      "componentId": "6851defe-87f9-41ec-bbda-4238a245b91a",
      "frameworkName": null,
      "frameworkSlug": null
    },
    {
      "bbox": null,
      "kind": "metric",
      "text": "Maximum Arena Score: 100-point",
      "attrs": null,
      "subkind": "primary",
      "toolName": "Quantification",
      "toolSlug": "quantification",
      "confidence": null,
      "componentId": "019dd952-41ac-7180-a821-a17910c42d90",
      "frameworkName": null,
      "frameworkSlug": null
    },
    {
      "bbox": {
        "h": 0.08,
        "w": 0.95,
        "x": 0.02,
        "y": 0.21
      },
      "kind": "paragraph",
      "text": "Researchers revealed systematic manipulation of LMArena as Meta tested 27 private Llama-4 variants before cherry-picking the winner: testing 10 model variants yields a 100-point boost.",
      "attrs": null,
      "subkind": "paragraph",
      "toolName": null,
      "toolSlug": null,
      "confidence": null,
      "componentId": "60c07b33-74cf-4762-9eeb-497fa6099126",
      "frameworkName": null,
      "frameworkSlug": null
    },
    {
      "bbox": {
        "h": 0.05,
        "w": 0.9,
        "x": 0.02,
        "y": 0.13
      },
      "kind": "title",
      "text": "The trouble with benchmarks: vibes aren't all we need, but increasingly all we've got",
      "attrs": null,
      "subkind": "headline",
      "toolName": null,
      "toolSlug": null,
      "confidence": null,
      "componentId": "33b27257-9acb-43ac-907c-468b6b0a268e",
      "frameworkName": null,
      "frameworkSlug": null
    }
  ],
  "metrics": [],
  "tools": [
    {
      "name": "Action Titles",
      "slug": "action-titles",
      "agent": "Architect",
      "layer": "slide",
      "matchId": "019dd95a-1055-74e0-a95c-1c689b0dd8e0",
      "evidence": "Title quotes 'vibes aren't all we need, but increasingly all we've got'.",
      "confidence": 85
    },
    {
      "name": "Concrete Language",
      "slug": "concrete-language",
      "agent": "Storyteller",
      "layer": "slide",
      "matchId": "019dd95a-1055-74e0-a95c-2189ae98402c",
      "evidence": "27 private Llama-4 variants → 100-point boost is concrete proof.",
      "confidence": 80
    }
  ],
  "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": [],
  "imagePathAlt": null,
  "thumbSrc": null,
  "thumbSrcAlt": null,
  "locked": true
}