{
  "kind": "framework",
  "value": "reinforcement-learning-from-human-feedback-rlhf-",
  "collectionKey": "slides:framework:reinforcement-learning-from-human-feedback-rlhf-:all-document-kinds:all-producers:all-orientations",
  "filters": {
    "documentKinds": [],
    "sourceTypes": [],
    "orientations": []
  },
  "total": 2,
  "page": 1,
  "pageSize": 24,
  "pageCount": 1,
  "rows": [
    {
      "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": 101,
      "slideType": "diagnosis",
      "function": "present_framework",
      "notes": "Includes a process diagram for RLHF and a bar chart comparing model performance.",
      "imagePath": null,
      "matchCount": 1,
      "evidence": "Explicitly mentioned in title and body text.",
      "slideHref": "/slides/019dd923-5e88-73ef-bd5d-01cf0d8a8fbc/101",
      "deckHref": "/decks/019dd923-5e88-73ef-bd5d-01cf0d8a8fbc",
      "deckAnchorHref": "/decks/019dd923-5e88-73ef-bd5d-01cf0d8a8fbc#slide-101",
      "loopMatches": [
        {
          "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"
        }
      ],
      "arcBeatMatches": [
        {
          "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"
        }
      ],
      "imagePathAlt": null,
      "thumbSrc": null,
      "thumbSrcAlt": null,
      "locked": true
    },
    {
      "docId": "019de06f-84e4-70e4-bce4-3c7c4f09a8ea",
      "docSlug": "602cc25b036280012a1a042017f6a35d",
      "documentTitle": "OpenAI | Product Presentation Deck | 20 slides",
      "authorId": "openai",
      "authorName": "OpenAI",
      "documentKindSlug": "conference-presentation",
      "documentKindLabel": "Conference presentation",
      "sourceTypeSlug": "investor_relations",
      "sourceTypeLabel": "Investor relations",
      "presentationDate": "2017-08-01 00:00:00",
      "orientation": "landscape",
      "aspectRatio": 1.7977529,
      "pageNumber": 11,
      "slideType": "propose_solution",
      "function": "present_framework",
      "notes": "The diagram shows a standard RL loop (RL Algorithm <-> Environment) augmented with a Reward Predictor that receives human feedback.",
      "imagePath": null,
      "matchCount": 1,
      "evidence": "Diagram shows RL algorithm, environment, and reward predictor with human feedback.",
      "slideHref": "/slides/019de06f-84e4-70e4-bce4-3c7c4f09a8ea/11",
      "deckHref": "/decks/019de06f-84e4-70e4-bce4-3c7c4f09a8ea",
      "deckAnchorHref": "/decks/019de06f-84e4-70e4-bce4-3c7c4f09a8ea#slide-11",
      "loopMatches": [
        {
          "to": 14,
          "from": 8,
          "name": "Jobs To Be Done",
          "slug": "53-jobs-to-be-done",
          "bestFor": "Product innovation, market entry, competitive positioning",
          "matchId": "fdcd1287-c4f6-4816-863c-2281b6ffccea",
          "evidence": "The deck proposes a solution using human feedback training processes and provides evidence to support it.",
          "position": 1,
          "objective": "How can we train an ML system to align with human preferences and values?",
          "structure": "The Customer's Job -> Current Solutions (Hired/Fired) -> Unmet Needs -> Our Solution Fit",
          "confidence": 0.7,
          "description": "Reframe the problem around what the customer is trying to accomplish, not what they're buying"
        }
      ],
      "arcBeatMatches": [
        {
          "to": 11,
          "from": 8,
          "beatId": "72a1fd78-bc62-4d21-a603-262dd269d030",
          "arcName": "Problem-Agitate-Solution",
          "arcSlug": "problem-agitate-solution",
          "beatName": "Solution",
          "beatSlug": "problem-agitate-solution-solution-provide-relief",
          "evidence": "The deck proposes a solution using human feedback training processes and provides evidence to support it.",
          "position": 2,
          "confidence": 0.9,
          "parentBeatName": "Resolution",
          "parentBeatSlug": "resolution"
        }
      ],
      "imagePathAlt": null,
      "thumbSrc": null,
      "thumbSrcAlt": null,
      "locked": true
    }
  ]
}