{
  "docId": "019dd923-5e88-73ef-bd5d-0856e1444fb9",
  "docSlug": "8757f1b44ef7f176",
  "documentTitle": "2024 Air Street Capital The State of AI Report 2024",
  "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": 184,
  "pageCount": 213,
  "prevPage": 183,
  "nextPage": 185,
  "slideType": "industry_trends",
  "function": "present_framework",
  "density": "balanced",
  "nDataPoints": 0,
  "notes": "The slide describes two specific automated red teaming methodologies: Rainbow Teaming and AdvPrompter.",
  "elementsJson": [
    "headline_text",
    "bullet_list",
    "process_diagram"
  ],
  "metadataConfidence": 1,
  "imagePath": null,
  "slideHref": "/slides/019dd923-5e88-73ef-bd5d-0856e1444fb9/184",
  "deckHref": "/decks/019dd923-5e88-73ef-bd5d-0856e1444fb9",
  "deckJsonHref": "/decks/019dd923-5e88-73ef-bd5d-0856e1444fb9.json",
  "deckAnchorHref": "/decks/019dd923-5e88-73ef-bd5d-0856e1444fb9#slide-184",
  "components": [
    {
      "bbox": null,
      "kind": "callout",
      "text": "Labs are increasingly using LLMs to scale the process of finding and patching vulnerabilities, including two teams at Meta.",
      "attrs": null,
      "subkind": null,
      "toolName": "Visual emphasis",
      "toolSlug": "visual-emphasis",
      "confidence": null,
      "componentId": "019dd952-47c3-7407-8d1f-105763fff69c",
      "frameworkName": null,
      "frameworkSlug": null
    },
    {
      "bbox": {
        "h": 0.45,
        "w": 0.48,
        "x": 0.5,
        "y": 0.45
      },
      "kind": "diagram",
      "text": "Automated red teaming process flow diagram showing interaction between Judge LLM, Target LLM, and Mutator LLM.",
      "attrs": null,
      "subkind": "process",
      "toolName": null,
      "toolSlug": null,
      "confidence": null,
      "componentId": "b5151624-9b31-4cad-8fe1-918a3b7274f7",
      "frameworkName": null,
      "frameworkSlug": null
    },
    {
      "bbox": {
        "h": 0.55,
        "w": 0.45,
        "x": 0.02,
        "y": 0.32
      },
      "kind": "list",
      "text": "Rainbow Teaming employs an open-ended search algorithm to create prompts that are designed to elicit potentially unsafe or biased responses from the target LLM.\nBy varying their approach and content, they can systematically explore LLM weaknesses. This was used as part of the safety testing for Llama 3.\nRather than evolutionary search, AdvPrompter uses a single LLM, going through an alternating process of generating adversarial prompts and fine-tuning on them.\nOnce trained, AdvPrompter can quickly produce new adversarial prompts adapted to different instructions.",
      "attrs": null,
      "subkind": "bullet",
      "toolName": null,
      "toolSlug": null,
      "confidence": null,
      "componentId": "4b6abe38-b1be-47e8-8049-2b4728b95a8a",
      "frameworkName": null,
      "frameworkSlug": null
    },
    {
      "bbox": {
        "h": 0.1,
        "w": 0.45,
        "x": 0.02,
        "y": 0.21
      },
      "kind": "paragraph",
      "text": "Coming up with endless potential attacks to red team models is challenging. Labs are increasingly using LLMs to scale the process of finding and patching vulnerabilities, including two teams at Meta.",
      "attrs": null,
      "subkind": "paragraph",
      "toolName": null,
      "toolSlug": null,
      "confidence": null,
      "componentId": "c790a438-966e-499f-a318-dc7e3825fe94",
      "frameworkName": null,
      "frameworkSlug": null
    },
    {
      "bbox": {
        "h": 0.04,
        "w": 0.45,
        "x": 0.02,
        "y": 0.14
      },
      "kind": "title",
      "text": "If you can't beat the jailbreakers, join them",
      "attrs": null,
      "subkind": "headline",
      "toolName": null,
      "toolSlug": null,
      "confidence": null,
      "componentId": "68a4f31e-c30f-4a73-a312-1ea15582ac70",
      "frameworkName": null,
      "frameworkSlug": null
    }
  ],
  "metrics": [],
  "tools": [
    {
      "name": "Action Titles",
      "slug": "action-titles",
      "agent": "Architect",
      "layer": "slide",
      "matchId": "019dd95a-1055-74e0-a95a-9f9799494a48",
      "evidence": "Title 'If you can't beat the jailbreakers, join them'",
      "confidence": 80
    },
    {
      "name": "Metaphor & Analogy",
      "slug": "metaphor-analogy",
      "agent": "Storyteller",
      "layer": "slide",
      "matchId": "019dd95a-1055-74e0-a95a-a1759396cbe1",
      "evidence": "Plays on 'if you can't beat them, join them' idiom",
      "confidence": 70
    }
  ],
  "frameworks": [
    {
      "name": "process",
      "slug": null,
      "matchId": "867078fa-0706-497d-ae44-3adac7bdef91",
      "evidence": "The diagram illustrates the iterative loop of automated red teaming.",
      "confidence": 0.8
    }
  ],
  "arcBeats": [
    {
      "to": 204,
      "from": 154,
      "beatId": "019dd95a-0682-776c-8e35-605159a069e6",
      "arcName": "The Triple Take",
      "arcSlug": "triple-take",
      "beatName": "The Implications (So What)",
      "beatSlug": "triple-take-the-implications-so-what",
      "evidence": "Politics + Safety sections explore consequences and risks",
      "position": 2,
      "confidence": 78,
      "parentBeatName": "Reflection",
      "parentBeatSlug": "reflection"
    },
    {
      "to": 204,
      "from": 175,
      "beatId": "019dd95a-0682-776c-8e35-74f05e198faf",
      "arcName": "Voyage and Return",
      "arcSlug": "voyage-return",
      "beatName": "The Return",
      "beatSlug": "voyage-return-the-return",
      "evidence": "Safety section returns to risks/governance debate",
      "position": 4,
      "confidence": 55,
      "parentBeatName": "Resolution",
      "parentBeatSlug": "resolution"
    }
  ],
  "loops": [
    {
      "to": 186,
      "from": 182,
      "name": "Pattern Hunter",
      "slug": "02-pattern-hunter",
      "bestFor": "Time-pressed audiences, building consensus, when data is strong",
      "matchId": "019dd95a-07fe-70ce-8d3e-76aa50b143e8",
      "evidence": "Attack surface, Pliny red team, LLM-on-LLM patching, adversarial robustness, stealth attacks",
      "position": 22,
      "objective": "Show jailbreaking arms race across attack surfaces",
      "structure": "Evidence A -> Evidence B -> Evidence C -> Pattern/Conclusion",
      "confidence": 75,
      "description": "Group multiple pieces of evidence that together point to a pattern or conclusion"
    }
  ],
  "imagePathAlt": null,
  "thumbSrc": null,
  "thumbSrcAlt": null,
  "locked": true
}