{
  "docId": "019dd923-5e88-73ef-bd5d-01cf0d8a8fbc",
  "docSlug": "cc8183ec02431b7a",
  "documentTitle": "2022 Air Street Capital The State of AI Report 2022",
  "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": 102,
  "pageCount": 114,
  "prevPage": 101,
  "nextPage": 103,
  "slideType": "industry_trends",
  "function": "illustrate_case",
  "density": "dense",
  "nDataPoints": 15,
  "notes": "Includes a conceptual diagram of the red teaming process and a bar chart showing harmlessness by model size.",
  "elementsJson": [
    "paragraph",
    "bullet_list",
    "bar_chart_grouped",
    "process_diagram"
  ],
  "metadataConfidence": 1,
  "imagePath": null,
  "slideHref": "/slides/019dd923-5e88-73ef-bd5d-01cf0d8a8fbc/102",
  "deckHref": "/decks/019dd923-5e88-73ef-bd5d-01cf0d8a8fbc",
  "deckJsonHref": "/decks/019dd923-5e88-73ef-bd5d-01cf0d8a8fbc.json",
  "deckAnchorHref": "/decks/019dd923-5e88-73ef-bd5d-01cf0d8a8fbc#slide-102",
  "components": [
    {
      "bbox": null,
      "kind": "callout",
      "text": "Anthropic used manual red teaming to evaluate RLHF models, finding that they are harder to attack and less harmful with increased model size.",
      "attrs": null,
      "subkind": null,
      "toolName": "Visual emphasis",
      "toolSlug": "visual-emphasis",
      "confidence": null,
      "componentId": "019dd952-47c4-7719-a21f-40877cb7bd15",
      "frameworkName": null,
      "frameworkSlug": null
    },
    {
      "bbox": {
        "h": 0.3,
        "w": 0.25,
        "x": 0.42,
        "y": 0.65
      },
      "kind": "chart",
      "text": "Average AI Harmlessness by Model Size",
      "attrs": null,
      "subkind": "bar-grouped",
      "toolName": null,
      "toolSlug": null,
      "confidence": null,
      "componentId": "9ca05bad-5891-4c4b-9f69-6019bbe202d3",
      "frameworkName": null,
      "frameworkSlug": null
    },
    {
      "bbox": {
        "h": 0.65,
        "w": 0.3,
        "x": 0.65,
        "y": 0.18
      },
      "kind": "diagram",
      "text": "Red Teaming workflow showing Red LM attacking Target LM with Red Clf evaluating output.",
      "attrs": null,
      "subkind": "process",
      "toolName": null,
      "toolSlug": null,
      "confidence": null,
      "componentId": "4bc0c0c1-f59d-47f1-91e0-0c9a97ce1160",
      "frameworkName": null,
      "frameworkSlug": null
    },
    {
      "bbox": {
        "h": 0.2,
        "w": 0.45,
        "x": 0.02,
        "y": 0.48
      },
      "kind": "list",
      "text": "Anthropic used manual red teaming to evaluate RLHF models, finding that they are harder to attack and less harmful with increased model size.\nIn the future, a classifier could detect for speculative risks such as power-seeking behavior or malicious coding.",
      "attrs": null,
      "subkind": "bullet",
      "toolName": null,
      "toolSlug": null,
      "confidence": null,
      "componentId": "2dd4dc31-d8d0-46b6-9e42-d97d39fac7b0",
      "frameworkName": null,
      "frameworkSlug": null
    },
    {
      "bbox": null,
      "kind": "metric",
      "text": "Average AI Harmlessness",
      "attrs": null,
      "subkind": "primary",
      "toolName": "Quantification",
      "toolSlug": "quantification",
      "confidence": null,
      "componentId": "019dd952-47c4-7719-a21f-45c2e82c3af4",
      "frameworkName": null,
      "frameworkSlug": null
    },
    {
      "bbox": {
        "h": 0.25,
        "w": 0.45,
        "x": 0.02,
        "y": 0.2
      },
      "kind": "paragraph",
      "text": "As language models exhibit an increasing array of capabilities, it becomes difficult to exhaustively evaluate their failure modes, inhibiting trust and safe public deployment. DeepMind introduced automated “red teaming”, in which manual testing can be complemented through using other language models to automatically “attack” other language models to make them exhibit unsafe behaviour, as determined by a separate classifier.",
      "attrs": null,
      "subkind": "paragraph",
      "toolName": null,
      "toolSlug": null,
      "confidence": null,
      "componentId": "43a26003-fd90-4814-b41e-6e80b56265b8",
      "frameworkName": null,
      "frameworkSlug": null
    },
    {
      "bbox": {
        "h": 0.04,
        "w": 0.4,
        "x": 0.02,
        "y": 0.14
      },
      "kind": "title",
      "text": "Language Model Alignment: Red Teaming",
      "attrs": null,
      "subkind": "headline",
      "toolName": null,
      "toolSlug": null,
      "confidence": null,
      "componentId": "12223ff8-6d75-4bb0-9742-850f85ee5b90",
      "frameworkName": null,
      "frameworkSlug": null
    }
  ],
  "metrics": [],
  "tools": [
    {
      "name": "Credibility Transfer",
      "slug": "credibility-transfer",
      "agent": "Storyteller",
      "layer": "slide",
      "matchId": "019dd95a-1055-74e0-a956-ca27a1b91418",
      "evidence": "Cites Anthropic's red-teaming work as authoritative.",
      "confidence": 78
    }
  ],
  "frameworks": [],
  "arcBeats": [
    {
      "to": 106,
      "from": 81,
      "beatId": "019dd95a-0682-776c-8e35-28fc5bded874",
      "arcName": "The Triple Take",
      "arcSlug": "triple-take",
      "beatName": "The Implications (So What)",
      "beatSlug": "triple-take-the-implications-so-what",
      "evidence": "Politics and Safety sections interpret what the facts mean for governance and risk.",
      "position": 2,
      "confidence": 78,
      "parentBeatName": "Reflection",
      "parentBeatSlug": "reflection"
    },
    {
      "to": 106,
      "from": 94,
      "beatId": "019dd95a-0682-776c-8e35-381428a808af",
      "arcName": "The Mountain",
      "arcSlug": "mountain",
      "beatName": "Climax",
      "beatSlug": "mountain-climax",
      "evidence": "Safety section confronts existential risk and alignment urgency.",
      "position": 3,
      "confidence": 55,
      "parentBeatName": "Turn",
      "parentBeatSlug": "turn"
    }
  ],
  "loops": [
    {
      "to": 106,
      "from": 100,
      "name": "Pattern Hunter",
      "slug": "02-pattern-hunter",
      "bestFor": "Time-pressed audiences, building consensus, when data is strong",
      "matchId": "019dd95a-07fe-70ce-8d3d-db1f31ffae02",
      "evidence": "RLHF, red-teaming, mechanistic interpretability, goal misgeneralization, moral behavior, Conjecture lab.",
      "position": 17,
      "objective": "Catalogue alignment-research techniques and example labs",
      "structure": "Evidence A -> Evidence B -> Evidence C -> Pattern/Conclusion",
      "confidence": 78,
      "description": "Group multiple pieces of evidence that together point to a pattern or conclusion"
    }
  ],
  "imagePathAlt": null,
  "thumbSrc": null,
  "thumbSrcAlt": null,
  "locked": true
}