{
  "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": 72,
  "pageCount": 313,
  "prevPage": 71,
  "nextPage": 73,
  "slideType": "diagnosis",
  "function": "diagnose",
  "density": "dense",
  "nDataPoints": 2,
  "notes": "The chart shows a negative correlation between capabilities score and attack failure rate, indicating that as models get more capable, they become more prone to failure in safety-critical tasks (WMDP).",
  "elementsJson": [
    "headline_text",
    "bullet_list",
    "scatter_plot",
    "callout_box"
  ],
  "metadataConfidence": 1,
  "imagePath": null,
  "slideHref": "/slides/019dd923-5e88-73ef-bd5d-0d0f98caffe1/72",
  "deckHref": "/decks/019dd923-5e88-73ef-bd5d-0d0f98caffe1",
  "deckJsonHref": "/decks/019dd923-5e88-73ef-bd5d-0d0f98caffe1.json",
  "deckAnchorHref": "/decks/019dd923-5e88-73ef-bd5d-0d0f98caffe1#slide-72",
  "components": [
    {
      "bbox": {
        "h": 0.08,
        "w": 0.95,
        "x": 0.02,
        "y": 0.22
      },
      "kind": "callout",
      "text": "71% of safety benchmark variance is explained by general capabilities alone, while genuine risks like WMDP bioweapons (-0.91) worsen as models get smarter.",
      "attrs": null,
      "subkind": "primary",
      "toolName": null,
      "toolSlug": null,
      "confidence": null,
      "componentId": "b4477fbd-73f8-41bd-94b1-34e030b0991a",
      "frameworkName": null,
      "frameworkSlug": null
    },
    {
      "bbox": {
        "h": 0.55,
        "w": 0.45,
        "x": 0.52,
        "y": 0.32
      },
      "kind": "chart",
      "text": "WMDP (-0.86) scatter plot showing Accuracy vs Capabilities score",
      "attrs": null,
      "subkind": "scatter",
      "toolName": null,
      "toolSlug": null,
      "confidence": null,
      "componentId": "062c3f1b-f71f-4689-a2d1-a84422c64859",
      "frameworkName": null,
      "frameworkSlug": null
    },
    {
      "bbox": {
        "h": 0.55,
        "w": 0.45,
        "x": 0.04,
        "y": 0.32
      },
      "kind": "list",
      "text": "Safety benchmarks are highly correlated with capabilities; simply scaling models improves most safety metrics! Yet critical safety issues worsen with scale: the most dangerous capabilities are inversely correlated with reported safety improvements. Instruction tuning masks rather than solves problems. Base model correlations flip from negative to positive after chat tuning (CyberSecEval: -0.25 → 0.55) while harmful capabilities remain latent in the model. Safety research should prioritise metrics which are not highly correlated with scale!",
      "attrs": null,
      "subkind": "bullet",
      "toolName": null,
      "toolSlug": null,
      "confidence": null,
      "componentId": "9ae7f625-18ac-455e-b991-134ad1a17e8b",
      "frameworkName": null,
      "frameworkSlug": null
    },
    {
      "bbox": null,
      "kind": "metric",
      "text": "Attack failure rate: 71%",
      "attrs": null,
      "subkind": "primary",
      "toolName": "Quantification",
      "toolSlug": "quantification",
      "confidence": null,
      "componentId": "019dd952-41ac-7180-a825-d41c18c3068f",
      "frameworkName": null,
      "frameworkSlug": null
    },
    {
      "bbox": {
        "h": 0.05,
        "w": 0.45,
        "x": 0.02,
        "y": 0.14
      },
      "kind": "title",
      "text": "The trouble with benchmarks: safety-washing",
      "attrs": null,
      "subkind": "headline",
      "toolName": null,
      "toolSlug": null,
      "confidence": null,
      "componentId": "563cb924-73d3-455f-bd8c-d0cf49073be3",
      "frameworkName": null,
      "frameworkSlug": null
    }
  ],
  "metrics": [],
  "tools": [
    {
      "name": "Action Titles",
      "slug": "action-titles",
      "agent": "Architect",
      "layer": "slide",
      "matchId": "019dd95a-1055-74e0-a95c-26b6f76c0abf",
      "evidence": "Title 'safety-washing' delivers the diagnosis.",
      "confidence": 85
    },
    {
      "name": "Anchoring Bias",
      "slug": "anchoring-bias",
      "agent": "Architect",
      "layer": "slide",
      "matchId": "f865712b-9fad-419e-b18c-69469dedc199",
      "evidence": "71% of safety benchmark variance is explained by general capabilities alone",
      "confidence": 0.7
    },
    {
      "name": "Concrete Language",
      "slug": "concrete-language",
      "agent": "Storyteller",
      "layer": "slide",
      "matchId": "019dd95a-1055-74e0-a95c-296bd2ebc593",
      "evidence": "'71% of safety benchmark variance explained by general capabilities'.",
      "confidence": 85
    }
  ],
  "frameworks": [
    {
      "name": "diagnostic-test",
      "slug": null,
      "matchId": "975632af-0419-46c4-a7ae-b8744761078c",
      "evidence": "The slide identifies a systemic flaw in current safety benchmarking practices.",
      "confidence": 0.8
    }
  ],
  "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
}