{ "schemaVersion": "deck-intelligence.v1", "id": "019de06f-84e4-70e4-bce4-3c7c4f09a8ea", "slug": "602cc25b036280012a1a042017f6a35d", "title": "OpenAI | Product Presentation Deck | 20 slides", "subtitle": "OpenAI · 2017-08", "author": "OpenAI", "date": "2017-08-01 00:00:00", "pageCount": 20, "kind": "conference-presentation", "orientation": "landscape", "aspectRatio": 1.7977529, "document": { "id": "019de06f-84e4-70e4-bce4-3c7c4f09a8ea", "slug": "602cc25b036280012a1a042017f6a35d", "title": "OpenAI | Product Presentation Deck | 20 slides", "rawTitle": "OpenAI | Product Presentation Deck | 20 slides", "authorId": "openai", "authorName": "OpenAI", "authorDisplay": "OpenAI", "kind": { "slug": "conference-presentation", "label": "Conference presentation" }, "sourceType": { "slug": "investor_relations", "label": "Investor relations" }, "fileName": "2017-08__openai__product__602cc25b.pdf", "sourceUrl": "https://www.slidebook.io/company/openai/presentation/602cc25b036280012a1a042017f6a35d/", "sourceDate": "2017-08-22 00:00:00", "presentationDate": "2017-08-01 00:00:00", "ingestedAt": "2026-04-30 22:08:18.252821+00", "status": "ready", "pageCount": 20, "orientation": "landscape", "aspectRatio": 1.7977529, "targetCompany": null, "targetTicker": null, "metadata": { "corpus": "slidebook", "industry": "information-technology-services", "local_root": "/Users/miguelsuredasuau/Desktop/lalalalla/_scrape/slidebook/decks/2017-08__openai__product__602cc25b", "image_width": 1920, "image_format": "jpg", "deck_type_raw": "Product Presentation Deck", "deck_type_slug": null, "source_url_pattern": "https://www.slidebook.io/company/{company_slug}/presentation/{deck_id}/" } }, "coverage": { "pageCount": 20, "pagesIndexed": 20, "pagesWithImages": 20, "pagesWithMetadata": 20, "imageCoveragePct": 100, "metadataCoveragePct": 100, "componentCount": 64, "chartCount": 3, "metricSlides": 0, "toolMatches": 8, "documentToolMatches": 0, "frameworkMatches": 7, "arcMatches": 1, "beatMatches": 3, "loopMatches": 2, "patternMatches": 0, "storymakersMatch": false, "scoreAvailable": false, "reviewAvailable": false, "activistThesisAvailable": false, "pitchdeckMetadataAvailable": false, "pitchdeckProfileAvailable": false }, "blocks": [], "matches": { "arcs": [ { "arc": { "name": "Problem-Agitate-Solution", "slug": "problem-agitate-solution", "status": "active", "bestFor": "Sales pitches, urgency-driven, competitive displacement", "canonId": "019dd956-5ec6-70f9-9fd8-fdbaaff39acf", "version": 1, "bodyDocId": "019df22a-2356-73eb-8061-d8c81dc0aa49", "structure": "Problem (Identify pain) -> Agitate (Make it worse) -> Solution (Provide relief)", "description": null, "familyLabel": null, "categoryName": "Strategy & Business", "categorySlug": "strategy-business" }, "matchId": "04c30632-8456-4dc0-8a7a-f7d40f1c761f", "evidence": "The deck follows a clear problem-agitate-solution structure, starting with a problem statement, highlighting the risks, and proposing a solution.", "isPrimary": true, "confidence": 0.9, "extraction": { "at": "2026-07-16 22:06:19.030437+00", "model": "or:meta-llama/llama-4-scout", "runId": null, "seconds": null, "promptVersion": "doc-narrative-v1" } } ], "beats": [ { "to": 4, "arc": { "name": "Problem-Agitate-Solution", "slug": "problem-agitate-solution", "status": "active", "bestFor": "Sales pitches, urgency-driven, competitive displacement", "canonId": "019dd956-5ec6-70f9-9fd8-fdbaaff39acf", "version": 1, "bodyDocId": "019df22a-2356-73eb-8061-d8c81dc0aa49", "structure": "Problem (Identify pain) -> Agitate (Make it worse) -> Solution (Provide relief)", "description": null, "familyLabel": null, "categoryName": "Strategy & Business", "categorySlug": "strategy-business" }, "from": 2, "name": "Problem", "beatId": "625ee8f7-4f93-455d-b525-f9a485b0c7c6", "matchId": "04c30632-8456-4dc0-8a7a-f7d40f1c761f", "beatType": { "name": "Problem (Identify pain)", "slug": "problem-agitate-solution-problem-identify-pain", "status": "active", "canonId": "019dd9b8-0e78-74bb-b8f6-54348367a131", "version": 1, "description": null }, "evidence": "The deck identifies potential negative social impacts of AI and sources of accident risk.", "position": 0, "isPrimaryArc": true, "parentBeatType": { "name": "Complication", "slug": "complication", "status": "active", "canonId": "019dd9b8-0275-7709-b2d6-4f77c3fc7579", "version": 1, "description": "The point in the narrative where the audience should feel discomfort. Required for argumentative arcs." }, "alignedBlockIds": null, "matchConfidence": 0.9 }, { "to": 6, "arc": { "name": "Problem-Agitate-Solution", "slug": "problem-agitate-solution", "status": "active", "bestFor": "Sales pitches, urgency-driven, competitive displacement", "canonId": "019dd956-5ec6-70f9-9fd8-fdbaaff39acf", "version": 1, "bodyDocId": "019df22a-2356-73eb-8061-d8c81dc0aa49", "structure": "Problem (Identify pain) -> Agitate (Make it worse) -> Solution (Provide relief)", "description": null, "familyLabel": null, "categoryName": "Strategy & Business", "categorySlug": "strategy-business" }, "from": 5, "name": "Agitate", "beatId": "0c5776c9-81ed-47ea-a4d1-51709ebdb4ed", "matchId": "04c30632-8456-4dc0-8a7a-f7d40f1c761f", "beatType": { "name": "Agitate (Make it worse)", "slug": "problem-agitate-solution-agitate-make-it-worse", "status": "active", "canonId": "019dd9b8-0f5a-71dd-bae7-8fadb0d09456", "version": 1, "description": null }, "evidence": "The deck highlights the risks of distributional shift and bad objective functions.", "position": 1, "isPrimaryArc": true, "parentBeatType": { "name": "Development", "slug": "development", "status": "active", "canonId": "019dd9b8-0332-7008-b3aa-11b0840a57b1", "version": 1, "description": "The middle stretch in arcs that have one. Often where length problems creep in." }, "alignedBlockIds": null, "matchConfidence": 0.9 }, { "to": 11, "arc": { "name": "Problem-Agitate-Solution", "slug": "problem-agitate-solution", "status": "active", "bestFor": "Sales pitches, urgency-driven, competitive displacement", "canonId": "019dd956-5ec6-70f9-9fd8-fdbaaff39acf", "version": 1, "bodyDocId": "019df22a-2356-73eb-8061-d8c81dc0aa49", "structure": "Problem (Identify pain) -> Agitate (Make it worse) -> Solution (Provide relief)", "description": null, "familyLabel": null, "categoryName": "Strategy & Business", "categorySlug": "strategy-business" }, "from": 8, "name": "Solution", "beatId": "72a1fd78-bc62-4d21-a603-262dd269d030", "matchId": "04c30632-8456-4dc0-8a7a-f7d40f1c761f", "beatType": { "name": "Solution (Provide relief)", "slug": "problem-agitate-solution-solution-provide-relief", "status": "active", "canonId": "019dd9b8-0fee-73e9-9ae5-799c7f3895cd", "version": 1, "description": null }, "evidence": "The deck proposes a solution using human feedback training processes and provides evidence to support it.", "position": 2, "isPrimaryArc": true, "parentBeatType": { "name": "Resolution", "slug": "resolution", "status": "active", "canonId": "019dd9b8-045d-703b-8dd7-e92aba3ba91b", "version": 1, "description": "How the deck ends. Required in 19/20 arcs. The exception (Sparkline) ends with new bliss which is functionally a resolution." }, "alignedBlockIds": null, "matchConfidence": 0.9 } ], "loops": [ { "to": 6, "from": 2, "loop": { "name": "27_cost_of_inaction", "slug": "27-cost-of-inaction", "status": "active", "bestFor": "Urgent budget requests, compliance, risk mitigation", "canonId": "019dd956-70d1-7395-a15b-857ba858b394", "version": 1, "bodyDocId": "019df22a-2420-77be-bc41-ded96d08cb21", "structure": "The Status Quo -> The Hidden Costs Accumulating -> The Future State of Inaction -> The Tipping Point", "description": "Quantify what happens if the audience does nothing", "familyLabel": null, "categoryName": "Urgency", "categorySlug": "urgency" }, "matchId": "17544e50-3a29-4ebc-91be-39b72c52ee43", "evidence": "The deck highlights the potential negative social impacts of AI and the risks of not addressing them.", "position": 0, "objective": "What are the risks of not addressing AI safety?", "confidence": 0.8, "extraction": { "at": "2026-07-16 22:06:19.16729+00", "model": "or:meta-llama/llama-4-scout", "runId": null, "seconds": null, "promptVersion": "doc-narrative-v1" } }, { "to": 14, "from": 8, "loop": { "name": "53_jobs_to_be_done", "slug": "53-jobs-to-be-done", "status": "active", "bestFor": "Product innovation, market entry, competitive positioning", "canonId": "019dd956-7b6c-7594-b5b3-eecbe0ab48e8", "version": 1, "bodyDocId": "019df22a-2420-77be-bc41-ded96d08cb21", "structure": "The Customer's Job -> Current Solutions (Hired/Fired) -> Unmet Needs -> Our Solution Fit", "description": "Reframe the problem around what the customer is trying to accomplish, not what they're buying", "familyLabel": null, "categoryName": "Consulting Framework", "categorySlug": "consulting-framework" }, "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?", "confidence": 0.7, "extraction": { "at": "2026-07-16 22:06:19.201225+00", "model": "or:meta-llama/llama-4-scout", "runId": null, "seconds": null, "promptVersion": "doc-narrative-v1" } } ], "tools": [ { "tool": { "name": "Problem Statement Canvas", "slug": "problem-statement-canvas", "status": "active", "bestFor": null, "canonId": "019dd956-839d-744b-ad11-9bf3e3642a18", "version": 1, "bodyDocId": null, "description": "Structured definition of the problem: What? Who? Where? When? Scale?", "familyLabel": null, "categoryName": "Block", "categorySlug": "block" }, "agent": "Architect", "layer": "slide", "agents": [ "Architect" ], "matchId": "3c07717d-3865-4b2a-ac40-cef8892fb58d", "evidence": "list/bullet: Misuse: Drone weapons, Mass surveillance, Advanced hacking, Military arms races. Social Side Effects: Unemployment, AI and criminal justice, Weaponized fake news.", "pageRefs": null, "priority": null, "whenToUse": null, "confidence": 0.7, "extraction": { "at": "2026-07-16 23:23:08.55679+00", "model": "or:meta-llama/llama-4-scout", "runId": null, "seconds": null, "promptVersion": null }, "pageNumber": 2, "whyItWorks": null, "antipattern": null, "cardinality": null, "narrativePurpose": null }, { "tool": { "name": "Issue Tree", "slug": "issue-tree", "status": "active", "bestFor": null, "canonId": "019dd956-82e3-775d-a2b6-318154564b53", "version": 1, "bodyDocId": null, "description": "Breaking down a problem into component parts systematically", "familyLabel": null, "categoryName": "Block", "categorySlug": "block" }, "agent": "Architect", "layer": "slide", "agents": [ "Architect" ], "matchId": "6f9741ac-f9d3-4c1a-94cd-c13134720671", "evidence": "list/bullet: Causes: A. Wrong objective function B. Right objective function, problem in learning or inference C. (Non-ML problems -- e.g. software implementation bug)", "pageRefs": null, "priority": null, "whenToUse": null, "confidence": 0.7, "extraction": { "at": "2026-07-16 23:23:08.669463+00", "model": "or:meta-llama/llama-4-scout", "runId": null, "seconds": null, "promptVersion": null }, "pageNumber": 4, "whyItWorks": null, "antipattern": null, "cardinality": null, "narrativePurpose": null }, { "tool": { "name": "Before-After-Bridge", "slug": "before-after-bridge", "status": "active", "bestFor": null, "canonId": "019dd956-8604-7188-883f-42370288ad05", "version": 1, "bodyDocId": null, "description": "Narrative structure: Paint the before, show the after, explain the bridge", "familyLabel": null, "categoryName": "Block", "categorySlug": "block" }, "agent": "Storyteller", "layer": "slide", "agents": [ "Storyteller" ], "matchId": "8b7db23b-7b58-4c89-b70d-814fe4487ec1", "evidence": "image/screenshot: CNET article screenshot: Google apologizes for algorithm mistakenly calling black people 'gorillas'", "pageRefs": null, "priority": null, "whenToUse": null, "confidence": 0.7, "extraction": { "at": "2026-07-16 23:23:08.782101+00", "model": "or:meta-llama/llama-4-scout", "runId": null, "seconds": null, "promptVersion": null }, "pageNumber": 6, "whyItWorks": null, "antipattern": null, "cardinality": null, "narrativePurpose": null }, { "tool": { "name": "Goodhart's Law", "slug": "goodharts-law", "status": "active", "bestFor": null, "canonId": "019dd956-8ed1-7421-ab8c-49dbce192ef5", "version": 1, "bodyDocId": null, "description": "When a measure becomes a target it ceases to be a good measure", "familyLabel": null, "categoryName": "Block", "categorySlug": "block" }, "agent": "Architect", "layer": "slide", "agents": [ "Architect" ], "matchId": "1ce75a51-24f8-4120-add4-501fd9568a00", "evidence": "Goodhart's law: \"When a metric becomes a target, it ceases to be a good", "pageRefs": null, "priority": null, "whenToUse": null, "confidence": 0.9, "extraction": { "at": "2026-07-16 23:23:08.896334+00", "model": "or:meta-llama/llama-4-scout", "runId": null, "seconds": null, "promptVersion": null }, "pageNumber": 7, "whyItWorks": null, "antipattern": null, "cardinality": null, "narrativePurpose": null }, { "tool": { "name": "List presentation", "slug": "list-presentation", "status": "active", "bestFor": null, "canonId": "019dd9e1-4ef7-777d-ba68-9074c2b3bcf1", "version": 1, "bodyDocId": null, "description": "Bulleted, numbered, or checklist enumerations.", "familyLabel": null, "categoryName": null, "categorySlug": null }, "agent": null, "layer": "slide", "agents": null, "matchId": "8292e732-8e47-41f6-91c9-c3812e4d2ad4", "evidence": "list/bullet: We currently address this by finding simple proxies for these complex goals, but this can lead to unexpected and in fact very unsafe behavior.", "pageRefs": null, "priority": null, "whenToUse": null, "confidence": 0.8, "extraction": { "at": "2026-07-16 23:23:08.857469+00", "model": "or:meta-llama/llama-4-scout", "runId": null, "seconds": null, "promptVersion": null }, "pageNumber": 7, "whyItWorks": null, "antipattern": null, "cardinality": "common", "narrativePurpose": null }, { "tool": { "name": "Process flow", "slug": "process-flow", "status": "active", "bestFor": null, "canonId": "019dd9e1-4dca-76bc-8135-401836a1c4c0", "version": 1, "bodyDocId": null, "description": "Workflow, steps, sequential diagram.", "familyLabel": null, "categoryName": null, "categorySlug": null }, "agent": null, "layer": "slide", "agents": null, "matchId": "9f7c2a90-5eab-477b-aa4d-36f5694096eb", "evidence": "image/screenshot: Two side-by-side simulation windows with 'Left is better' and 'Right is better' buttons above them.", "pageRefs": null, "priority": null, "whenToUse": null, "confidence": 0.7, "extraction": { "at": "2026-07-16 23:23:09.050328+00", "model": "or:meta-llama/llama-4-scout", "runId": null, "seconds": null, "promptVersion": null }, "pageNumber": 10, "whyItWorks": null, "antipattern": null, "cardinality": "common", "narrativePurpose": null }, { "tool": { "name": "Before-After-Bridge", "slug": "before-after-bridge", "status": "active", "bestFor": null, "canonId": "019dd956-8604-7188-883f-42370288ad05", "version": 1, "bodyDocId": null, "description": "Narrative structure: Paint the before, show the after, explain the bridge", "familyLabel": null, "categoryName": "Block", "categorySlug": "block" }, "agent": "Storyteller", "layer": "slide", "agents": [ "Storyteller" ], "matchId": "47e48a5c-c1cf-4d0a-8da6-32c8a3163bdd", "evidence": "The slide shows two gameplay screenshots with a headline 'Training 2 different behaviors on the same game'.", "pageRefs": null, "priority": null, "whenToUse": null, "confidence": 0.8, "extraction": { "at": "2026-07-16 23:22:57.988001+00", "model": "or:meta-llama/llama-4-scout", "runId": null, "seconds": null, "promptVersion": null }, "pageNumber": 12, "whyItWorks": null, "antipattern": null, "cardinality": null, "narrativePurpose": null }, { "tool": { "name": "Storytelling: the closing ask", "slug": "closing-ask", "status": "active", "bestFor": null, "canonId": "019df269-24e0-70ab-8dd9-983832c1c665", "version": 1, "bodyDocId": "019df269-24bc-7554-b88d-245127606b8d", "description": "The single most important slide in the deck — after the cover. The final", "familyLabel": "narrative-block", "categoryName": "Narrative & Transformation", "categorySlug": "narrative-transformation" }, "agent": "storyteller", "layer": "slide", "agents": [ "storyteller" ], "matchId": "e9be09aa-a02d-48c3-9424-2002a53ebf4e", "evidence": "The slide has a headline 'Q & A'.", "pageRefs": null, "priority": null, "whenToUse": null, "confidence": 0.8, "extraction": { "at": "2026-07-16 23:22:58.270719+00", "model": "or:meta-llama/llama-4-scout", "runId": null, "seconds": null, "promptVersion": null }, "pageNumber": 20, "whyItWorks": null, "antipattern": null, "cardinality": null, "narrativePurpose": null } ], "frameworks": [ { "matchId": "0c07c704-2b3b-4527-9309-319ee0362e41", "evidence": "Explicitly cited in the final bullet point.", "framework": null, "confidence": 1, "extraction": { "at": "2026-05-02 14:21:43.605+00", "model": "google/gemini-3.1-flash-lite-preview-20260303", "runId": null, "seconds": null, "promptVersion": null }, "pageNumber": 7, "frameworkName": "goodharts-law" }, { "matchId": "5c3de249-d9fa-4e70-9b26-356d456b6e21", "evidence": "Citing a seminal academic paper to establish credibility.", "framework": null, "confidence": 0.9, "extraction": { "at": "2026-05-02 14:21:41.083+00", "model": "google/gemini-3.1-flash-lite-preview-20260303", "runId": null, "seconds": null, "promptVersion": null }, "pageNumber": 9, "frameworkName": "authority-citation" }, { "matchId": "19c5ff67-57ca-4a5d-ade2-2718efa28e53", "evidence": "The slide depicts the binary choice mechanism central to RLHF.", "framework": null, "confidence": 0.95, "extraction": { "at": "2026-05-02 14:21:43.585+00", "model": "google/gemini-3.1-flash-lite-preview-20260303", "runId": null, "seconds": null, "promptVersion": null }, "pageNumber": 10, "frameworkName": "reinforcement-learning-from-human-feedback" }, { "matchId": "11d8debb-9c89-4b9c-984f-2bdce2f78139", "evidence": "Diagram shows RL algorithm, environment, and reward predictor with human feedback.", "framework": null, "confidence": 1, "extraction": { "at": "2026-05-02 14:21:41.665+00", "model": "google/gemini-3.1-flash-lite-preview-20260303", "runId": null, "seconds": null, "promptVersion": null }, "pageNumber": 11, "frameworkName": "Reinforcement Learning from Human Feedback (RLHF)" }, { "matchId": "45f2b249-2575-42fd-9bcc-995502dd0c59", "evidence": "The slide presents two distinct outcomes (behaviors) from the same input (game) to illustrate a point.", "framework": null, "confidence": 0.8, "extraction": { "at": "2026-05-02 14:21:45.478+00", "model": "google/gemini-3.1-flash-lite-preview-20260303", "runId": null, "seconds": null, "promptVersion": null }, "pageNumber": 12, "frameworkName": "before-after-framing" }, { "matchId": "f9b7a512-2e79-4e13-849b-f1931b525a64", "evidence": "The diagram explicitly labels 'Agent' and 'Environment' with state transitions (ot, ot+1) and action (at).", "framework": null, "confidence": 0.95, "extraction": { "at": "2026-05-02 14:21:42.642+00", "model": "google/gemini-3.1-flash-lite-preview-20260303", "runId": null, "seconds": null, "promptVersion": null }, "pageNumber": 17, "frameworkName": "Reinforcement Learning Agent-Environment Loop" }, { "matchId": "d06b1117-3761-4f07-b12a-2d142e6faced", "evidence": "Uses a reputable media source to validate the premise of the problem.", "framework": null, "confidence": 0.8, "extraction": { "at": "2026-05-02 14:21:44.729+00", "model": "google/gemini-3.1-flash-lite-preview-20260303", "runId": null, "seconds": null, "promptVersion": null }, "pageNumber": 18, "frameworkName": "authority-citation" } ], "patterns": [] }, "storymakers": null, "score": null, "review": null, "activistThesis": null, "pitchdeck": { "metadata": null, "profile": null }, "slides": [ { "page": 1, "type": "setup", "function": "front_matter", "imagePath": "https://imgproxy.kitesheet.com/ZlxBThi8QKdcsIaMziDKpZwLCqsgLo9liO7JC1FAxSg/rs:fit:1200:1200:0/q:82/f:webp/czM6Ly9raXRlc2hlZXQvY29ycHVzL3NsaWRlcy82MDJjYzI1YjAzNjI4MDAxMmExYTA0MjAxN2Y2YTM1ZC9wMDAxLmpwZw", "rawType": "cover", "block": null, "metadata": { "slideType": "cover", "slideTypeCanon": { "name": "Cover", "slug": "cover", "status": "active", "canonId": "019de52d-05f2-729d-9f93-e0504f0a2410", "version": 1, "description": null }, "function": "front_matter", "functionCanon": { "name": "Front matter", "slug": "front_matter", "status": "active", "canonId": "019de52d-133d-73f0-9e26-201eda96b098", "version": 1, "description": null }, "density": "sparse", "densityScore": 12, "componentCount": 3, "textChars": 79, "nDataPoints": 0, "notes": null, "elementsJson": null, "confidence": 1, "extraction": { "model": "google/gemini-3.1-flash-lite-preview-20260303", "runId": null, "at": "2026-05-02 14:21:43.108+00", "seconds": 5.733, "promptVersion": "v2-canon-conditioned" } }, "links": { "slide": "/slides/019de06f-84e4-70e4-bce4-3c7c4f09a8ea/1", "deck": "/decks/019de06f-84e4-70e4-bce4-3c7c4f09a8ea", "deckAnchor": "/decks/019de06f-84e4-70e4-bce4-3c7c4f09a8ea#slide-1", "deckJson": "/decks/019de06f-84e4-70e4-bce4-3c7c4f09a8ea.json" }, "components": [ { "bbox": { "h": 0.2, "w": 0.3, "x": 0.35, "y": 0.05 }, "kind": "image", "text": "OpenAI", "attrs": null, "canon": { "kind": { "name": "Image", "slug": "image", "status": "active", "canonId": "019de52c-f978-701f-a720-51d029d769e7", "version": 1, "description": "Photo, illustration, screenshot or icon." }, "tool": null, "subkind": { "name": "Logo", "slug": "logo", "status": "active", "canonId": "019de52c-fdaf-743d-94c4-d7c6a0e55302", "version": 1, "description": null }, "framework": null }, "subkind": "logo", "groupRole": null, "confidence": null, "extraction": { "at": "2026-05-02 14:21:43.108+00", "model": "google/gemini-3.1-flash-lite-preview-20260303", "runId": null, "seconds": null, "promptVersion": null }, "componentId": "29a6183f-c2b1-48f1-82f3-b7e0125356b2", "parentComponentId": null }, { "bbox": { "h": 0.15, "w": 0.4, "x": 0.3, "y": 0.75 }, "kind": "paragraph", "text": "Dario Amodei\n8/22/2017", "attrs": null, "canon": { "kind": { "name": "Paragraph", "slug": "paragraph", "status": "active", "canonId": "5f2affc0-941d-4d07-959d-4bfc1930cea4", "version": 1, "description": "Single prose block in the body of a slide." }, "tool": null, "subkind": { "name": "Paragraph", "slug": "paragraph", "status": "active", "canonId": "019de52c-fb80-75af-aa5c-a7a4bfac2433", "version": 1, "description": null }, "framework": null }, "subkind": "paragraph", "groupRole": null, "confidence": null, "extraction": { "at": "2026-05-02 14:21:43.108+00", "model": "google/gemini-3.1-flash-lite-preview-20260303", "runId": null, "seconds": null, "promptVersion": null }, "componentId": "499165fd-5f2a-4860-8cb5-aee3d32c68bd", "parentComponentId": null }, { "bbox": { "h": 0.3, "w": 0.8, "x": 0.1, "y": 0.35 }, "kind": "title", "text": "Human Preference Learning as a Direction in Safe AI", "attrs": null, "canon": { "kind": { "name": "Title", "slug": "title", "status": "active", "canonId": "019de52c-f7dd-76d9-b848-b1f12bd85cf4", "version": 1, "description": "Slide title or headline." }, "tool": null, "subkind": { "name": "Headline", "slug": "headline", "status": "active", "canonId": "019de52c-fae1-7463-9a7c-d60668eced0d", "version": 1, "description": null }, "framework": null }, "subkind": "headline", "groupRole": null, "confidence": null, "extraction": { "at": "2026-05-02 14:21:43.108+00", "model": "google/gemini-3.1-flash-lite-preview-20260303", "runId": null, "seconds": null, "promptVersion": null }, "componentId": "0596d554-6f6b-4ca8-8740-bcb936dc9d2d", "parentComponentId": null } ], "metrics": [], "tools": [], "frameworks": [], "arcBeats": [], "loops": [], "thumbSrc": "https://imgproxy.kitesheet.com/-__VswtouzFrznYFGl2xO4akb2GXfOhpoLF-WQIpE3g/rs:fit:480:480:0/q:78/f:webp/czM6Ly9raXRlc2hlZXQvY29ycHVzL3NsaWRlcy82MDJjYzI1YjAzNjI4MDAxMmExYTA0MjAxN2Y2YTM1ZC9wMDAxLmpwZw", "imagePathAlt": "https://imgproxy.kitesheet.com/tdZhBQmzsv3MIFTHM4bTUemeE359bcNJbw6tO8UIyKA/rs:fit:1200:1200:0/q:82/f:webp/czM6Ly9raXRlc2hlZXQvY29ycHVzL3NsaWRlcy82MDJjYzI1YjAzNjI4MDAxMmExYTA0MjAxN2Y2YTM1ZC9wMDAxLnBuZw", "thumbSrcAlt": "https://imgproxy.kitesheet.com/UeqK2c1snKYLOh0ZfN482DgUbgaXVA423pdKpMgPG8g/rs:fit:480:480:0/q:78/f:webp/czM6Ly9raXRlc2hlZXQvY29ycHVzL3NsaWRlcy82MDJjYzI1YjAzNjI4MDAxMmExYTA0MjAxN2Y2YTM1ZC9wMDAxLnBuZw" }, { "page": 2, "type": "tension", "function": "frame_problem", "rawType": "problem_statement", "block": null, "metadata": { "slideType": "problem_statement", "slideTypeCanon": { "name": "Problem statement", "slug": "problem_statement", "status": "active", "canonId": "019de52d-0879-768d-bba8-8da8a5a121d4", "version": 1, "description": null }, "function": "frame_problem", "functionCanon": { "name": "Frame problem", "slug": "frame_problem", "status": "active", "canonId": "019de52d-13db-753e-ad70-e9b2d0dab398", "version": 1, "description": null }, "density": "sparse", "densityScore": 15, "componentCount": 2, "textChars": 201, "nDataPoints": 0, "notes": null, "elementsJson": null, "confidence": 1, "extraction": { "model": "google/gemini-3.1-flash-lite-preview-20260303", "runId": null, "at": "2026-05-02 14:21:39.693+00", "seconds": 2.318, "promptVersion": "v2-canon-conditioned" } }, "links": { "slide": "/slides/019de06f-84e4-70e4-bce4-3c7c4f09a8ea/2", "deck": "/decks/019de06f-84e4-70e4-bce4-3c7c4f09a8ea", "deckAnchor": "/decks/019de06f-84e4-70e4-bce4-3c7c4f09a8ea#slide-2", "deckJson": "/decks/019de06f-84e4-70e4-bce4-3c7c4f09a8ea.json" }, "components": [ { "bbox": { "h": 0.7, "w": 0.4, "x": 0.05, "y": 0.2 }, "kind": "list", "text": "Misuse: Drone weapons, Mass surveillance, Advanced hacking, Military arms races. Social Side Effects: Unemployment, AI and criminal justice, Weaponized fake news.", "attrs": null, "canon": { "kind": { "name": "List", "slug": "list", "status": "active", "canonId": "019de52c-f94b-779b-a53b-fd4904a5f4d2", "version": 1, "description": "Bullet or numbered list." }, "tool": null, "subkind": { "name": "Bullet list", "slug": "bullet", "status": "active", "canonId": "019de52c-fc98-7169-a083-5cab805076ca", "version": 1, "description": null }, "framework": null }, "subkind": "bullet", "groupRole": null, "confidence": null, "extraction": { "at": "2026-05-02 14:21:39.693+00", "model": "google/gemini-3.1-flash-lite-preview-20260303", "runId": null, "seconds": null, "promptVersion": null }, "componentId": "020770f5-d7ef-48c5-a386-80074d069966", "parentComponentId": null }, { "bbox": { "h": 0.08, "w": 0.5, "x": 0.05, "y": 0.08 }, "kind": "title", "text": "Potential Negative Social Impacts of AI", "attrs": null, "canon": { "kind": { "name": "Title", "slug": "title", "status": "active", "canonId": "019de52c-f7dd-76d9-b848-b1f12bd85cf4", "version": 1, "description": "Slide title or headline." }, "tool": null, "subkind": { "name": "Headline", "slug": "headline", "status": "active", "canonId": "019de52c-fae1-7463-9a7c-d60668eced0d", "version": 1, "description": null }, "framework": null }, "subkind": "headline", "groupRole": null, "confidence": null, "extraction": { "at": "2026-05-02 14:21:39.693+00", "model": "google/gemini-3.1-flash-lite-preview-20260303", "runId": null, "seconds": null, "promptVersion": null }, "componentId": "9b1aadf4-e111-4253-b506-0d1977b341d0", "parentComponentId": null } ], "metrics": [], "tools": [ { "tool": { "name": "Problem Statement Canvas", "slug": "problem-statement-canvas", "status": "active", "bestFor": null, "canonId": "019dd956-839d-744b-ad11-9bf3e3642a18", "version": 1, "bodyDocId": null, "description": "Structured definition of the problem: What? Who? Where? When? Scale?", "familyLabel": null, "categoryName": "Block", "categorySlug": "block" }, "agent": "Architect", "layer": "slide", "agents": [ "Architect" ], "matchId": "3c07717d-3865-4b2a-ac40-cef8892fb58d", "evidence": "list/bullet: Misuse: Drone weapons, Mass surveillance, Advanced hacking, Military arms races. Social Side Effects: Unemployment, AI and criminal justice, Weaponized fake news.", "pageRefs": null, "priority": null, "whenToUse": null, "confidence": 0.7, "extraction": { "at": "2026-07-16 23:23:08.55679+00", "model": "or:meta-llama/llama-4-scout", "runId": null, "seconds": null, "promptVersion": null }, "pageNumber": 2, "whyItWorks": null, "antipattern": null, "cardinality": null, "narrativePurpose": null } ], "frameworks": [], "arcBeats": [ { "to": 4, "arc": { "name": "Problem-Agitate-Solution", "slug": "problem-agitate-solution", "status": "active", "bestFor": "Sales pitches, urgency-driven, competitive displacement", "canonId": "019dd956-5ec6-70f9-9fd8-fdbaaff39acf", "version": 1, "bodyDocId": "019df22a-2356-73eb-8061-d8c81dc0aa49", "structure": "Problem (Identify pain) -> Agitate (Make it worse) -> Solution (Provide relief)", "description": null, "familyLabel": null }, "from": 2, "name": "Problem", "beatId": "625ee8f7-4f93-455d-b525-f9a485b0c7c6", "matchId": "04c30632-8456-4dc0-8a7a-f7d40f1c761f", "beatType": { "name": "Problem (Identify pain)", "slug": "problem-agitate-solution-problem-identify-pain", "status": "active", "canonId": "019dd9b8-0e78-74bb-b8f6-54348367a131", "version": 1, "description": null }, "evidence": "The deck identifies potential negative social impacts of AI and sources of accident risk.", "position": 0, "isPrimaryArc": true, "parentBeatType": { "name": "Complication", "slug": "complication", "status": "active", "canonId": "019dd9b8-0275-7709-b2d6-4f77c3fc7579", "version": 1, "description": "The point in the narrative where the audience should feel discomfort. Required for argumentative arcs." }, "alignedBlockIds": null, "matchConfidence": 0.9 } ], "loops": [ { "to": 6, "from": 2, "loop": { "name": "27_cost_of_inaction", "slug": "27-cost-of-inaction", "status": "active", "bestFor": "Urgent budget requests, compliance, risk mitigation", "canonId": "019dd956-70d1-7395-a15b-857ba858b394", "version": 1, "bodyDocId": "019df22a-2420-77be-bc41-ded96d08cb21", "structure": "The Status Quo -> The Hidden Costs Accumulating -> The Future State of Inaction -> The Tipping Point", "description": "Quantify what happens if the audience does nothing", "familyLabel": null }, "matchId": "17544e50-3a29-4ebc-91be-39b72c52ee43", "evidence": "The deck highlights the potential negative social impacts of AI and the risks of not addressing them.", "position": 0, "objective": "What are the risks of not addressing AI safety?", "confidence": 0.8, "extraction": { "at": "2026-07-16 22:06:19.16729+00", "model": "or:meta-llama/llama-4-scout", "runId": null, "seconds": null, "promptVersion": "doc-narrative-v1" } } ], "locked": true }, { "page": 3, "type": "setup", "title": "The slide displays the title, author list, and abstract of a well-known academic paper.", "function": "front_matter", "rawType": "cover", "block": null, "metadata": { "slideType": "cover", "slideTypeCanon": { "name": "Cover", "slug": "cover", "status": "active", "canonId": "019de52d-05f2-729d-9f93-e0504f0a2410", "version": 1, "description": null }, "function": "front_matter", "functionCanon": { "name": "Front matter", "slug": "front_matter", "status": "active", "canonId": "019de52d-133d-73f0-9e26-201eda96b098", "version": 1, "description": null }, "density": "balanced", "densityScore": 40, "componentCount": 4, "textChars": 386, "nDataPoints": 0, "notes": "The slide displays the title, author list, and abstract of a well-known academic paper.", "elementsJson": null, "confidence": 1, "extraction": { "model": "google/gemini-3.1-flash-lite-preview-20260303", "runId": null, "at": "2026-05-02 14:21:39.786+00", "seconds": 2.352, "promptVersion": "v2-canon-conditioned" } }, "links": { "slide": "/slides/019de06f-84e4-70e4-bce4-3c7c4f09a8ea/3", "deck": "/decks/019de06f-84e4-70e4-bce4-3c7c4f09a8ea", "deckAnchor": "/decks/019de06f-84e4-70e4-bce4-3c7c4f09a8ea#slide-3", "deckJson": "/decks/019de06f-84e4-70e4-bce4-3c7c4f09a8ea.json" }, "components": [ { "bbox": { "h": 0.3, "w": 0.6, "x": 0.2, "y": 0.5 }, "kind": "paragraph", "text": "Rapid progress in machine learning and artificial intelligence (AI) has brought increasing attention to the potential impacts of AI technologies on society. In this paper we discuss one such potential impact: the problem of accidents in machine learning systems...", "attrs": null, "canon": { "kind": { "name": "Paragraph", "slug": "paragraph", "status": "active", "canonId": "5f2affc0-941d-4d07-959d-4bfc1930cea4", "version": 1, "description": "Single prose block in the body of a slide." }, "tool": null, "subkind": { "name": "Paragraph", "slug": "paragraph", "status": "active", "canonId": "019de52c-fb80-75af-aa5c-a7a4bfac2433", "version": 1, "description": null }, "framework": null }, "subkind": "paragraph", "groupRole": null, "confidence": null, "extraction": { "at": "2026-05-02 14:21:39.786+00", "model": "google/gemini-3.1-flash-lite-preview-20260303", "runId": null, "seconds": null, "promptVersion": null }, "componentId": "13f06b84-7c57-42d6-afca-314087d5e905", "parentComponentId": null }, { "bbox": { "h": 0.1, "w": 0.6, "x": 0.2, "y": 0.25 }, "kind": "paragraph", "text": "Dario Amodei, Chris Olah, Jacob Steinhardt, Paul Christiano, John Schulman, Dan Mané", "attrs": null, "canon": { "kind": { "name": "Paragraph", "slug": "paragraph", "status": "active", "canonId": "5f2affc0-941d-4d07-959d-4bfc1930cea4", "version": 1, "description": "Single prose block in the body of a slide." }, "tool": null, "subkind": { "name": "Paragraph", "slug": "paragraph", "status": "active", "canonId": "019de52c-fb80-75af-aa5c-a7a4bfac2433", "version": 1, "description": null }, "framework": null }, "subkind": "paragraph", "groupRole": null, "confidence": null, "extraction": { "at": "2026-05-02 14:21:39.786+00", "model": "google/gemini-3.1-flash-lite-preview-20260303", "runId": null, "seconds": null, "promptVersion": null }, "componentId": "d1adaf96-29ba-41f9-a42d-387fcc4897d6", "parentComponentId": null }, { "bbox": { "h": 0.05, "w": 0.5, "x": 0.25, "y": 0.15 }, "kind": "title", "text": "Concrete Problems in AI Safety", "attrs": null, "canon": { "kind": { "name": "Title", "slug": "title", "status": "active", "canonId": "019de52c-f7dd-76d9-b848-b1f12bd85cf4", "version": 1, "description": "Slide title or headline." }, "tool": null, "subkind": { "name": "Headline", "slug": "headline", "status": "active", "canonId": "019de52c-fae1-7463-9a7c-d60668eced0d", "version": 1, "description": null }, "framework": null }, "subkind": "headline", "groupRole": null, "confidence": null, "extraction": { "at": "2026-05-02 14:21:39.786+00", "model": "google/gemini-3.1-flash-lite-preview-20260303", "runId": null, "seconds": null, "promptVersion": null }, "componentId": "b7d0253e-2608-4e5c-b02e-6478db2e2087", "parentComponentId": null }, { "bbox": { "h": 0.03, "w": 0.1, "x": 0.45, "y": 0.45 }, "kind": "title", "text": "Abstract", "attrs": null, "canon": { "kind": { "name": "Title", "slug": "title", "status": "active", "canonId": "019de52c-f7dd-76d9-b848-b1f12bd85cf4", "version": 1, "description": "Slide title or headline." }, "tool": null, "subkind": { "name": "Subtitle", "slug": "subtitle", "status": "active", "canonId": "019de52c-fb09-739d-900c-815b0d9ac526", "version": 1, "description": null }, "framework": null }, "subkind": "subtitle", "groupRole": null, "confidence": null, "extraction": { "at": "2026-05-02 14:21:39.786+00", "model": "google/gemini-3.1-flash-lite-preview-20260303", "runId": null, "seconds": null, "promptVersion": null }, "componentId": "b484fc58-e2a9-4449-bda3-51f4fae33de3", "parentComponentId": null } ], "metrics": [], "tools": [], "frameworks": [], "arcBeats": [ { "to": 4, "arc": { "name": "Problem-Agitate-Solution", "slug": "problem-agitate-solution", "status": "active", "bestFor": "Sales pitches, urgency-driven, competitive displacement", "canonId": "019dd956-5ec6-70f9-9fd8-fdbaaff39acf", "version": 1, "bodyDocId": "019df22a-2356-73eb-8061-d8c81dc0aa49", "structure": "Problem (Identify pain) -> Agitate (Make it worse) -> Solution (Provide relief)", "description": null, "familyLabel": null }, "from": 2, "name": "Problem", "beatId": "625ee8f7-4f93-455d-b525-f9a485b0c7c6", "matchId": "04c30632-8456-4dc0-8a7a-f7d40f1c761f", "beatType": { "name": "Problem (Identify pain)", "slug": "problem-agitate-solution-problem-identify-pain", "status": "active", "canonId": "019dd9b8-0e78-74bb-b8f6-54348367a131", "version": 1, "description": null }, "evidence": "The deck identifies potential negative social impacts of AI and sources of accident risk.", "position": 0, "isPrimaryArc": true, "parentBeatType": { "name": "Complication", "slug": "complication", "status": "active", "canonId": "019dd9b8-0275-7709-b2d6-4f77c3fc7579", "version": 1, "description": "The point in the narrative where the audience should feel discomfort. Required for argumentative arcs." }, "alignedBlockIds": null, "matchConfidence": 0.9 } ], "loops": [ { "to": 6, "from": 2, "loop": { "name": "27_cost_of_inaction", "slug": "27-cost-of-inaction", "status": "active", "bestFor": "Urgent budget requests, compliance, risk mitigation", "canonId": "019dd956-70d1-7395-a15b-857ba858b394", "version": 1, "bodyDocId": "019df22a-2420-77be-bc41-ded96d08cb21", "structure": "The Status Quo -> The Hidden Costs Accumulating -> The Future State of Inaction -> The Tipping Point", "description": "Quantify what happens if the audience does nothing", "familyLabel": null }, "matchId": "17544e50-3a29-4ebc-91be-39b72c52ee43", "evidence": "The deck highlights the potential negative social impacts of AI and the risks of not addressing them.", "position": 0, "objective": "What are the risks of not addressing AI safety?", "confidence": 0.8, "extraction": { "at": "2026-07-16 22:06:19.16729+00", "model": "or:meta-llama/llama-4-scout", "runId": null, "seconds": null, "promptVersion": "doc-narrative-v1" } } ], "locked": true }, { "page": 4, "type": "tension", "title": "The slide uses a simple list to categorize failure modes, with a specific exclusion for non-ML problems.", "function": "diagnose_problem", "rawType": "problem_statement", "block": null, "metadata": { "slideType": "problem_statement", "slideTypeCanon": { "name": "Problem statement", "slug": "problem_statement", "status": "active", "canonId": "019de52d-0879-768d-bba8-8da8a5a121d4", "version": 1, "description": null }, "function": "diagnose_problem", "functionCanon": { "name": "Diagnose problem", "slug": "diagnose_problem", "status": "active", "canonId": "019de52d-176d-761f-b608-cf52262d2ab2", "version": 1, "description": null }, "density": "balanced", "densityScore": 27, "componentCount": 3, "textChars": 316, "nDataPoints": 0, "notes": "The slide uses a simple list to categorize failure modes, with a specific exclusion for non-ML problems.", "elementsJson": null, "confidence": 1, "extraction": { "model": "google/gemini-3.1-flash-lite-preview-20260303", "runId": null, "at": "2026-05-02 14:21:43.374+00", "seconds": 5.854, "promptVersion": "v2-canon-conditioned" } }, "links": { "slide": "/slides/019de06f-84e4-70e4-bce4-3c7c4f09a8ea/4", "deck": "/decks/019de06f-84e4-70e4-bce4-3c7c4f09a8ea", "deckAnchor": "/decks/019de06f-84e4-70e4-bce4-3c7c4f09a8ea#slide-4", "deckJson": "/decks/019de06f-84e4-70e4-bce4-3c7c4f09a8ea.json" }, "components": [ { "bbox": { "h": 0.3, "w": 0.7, "x": 0.05, "y": 0.4 }, "kind": "list", "text": "Causes:\nA. Wrong objective function\nB. Right objective function, problem in learning or inference\nC. (Non-ML problems -- e.g. software implementation bug)", "attrs": null, "canon": { "kind": { "name": "List", "slug": "list", "status": "active", "canonId": "019de52c-f94b-779b-a53b-fd4904a5f4d2", "version": 1, "description": "Bullet or numbered list." }, "tool": null, "subkind": { "name": "Bullet list", "slug": "bullet", "status": "active", "canonId": "019de52c-fc98-7169-a083-5cab805076ca", "version": 1, "description": null }, "framework": null }, "subkind": "bullet", "groupRole": null, "confidence": null, "extraction": { "at": "2026-05-02 14:21:43.374+00", "model": "google/gemini-3.1-flash-lite-preview-20260303", "runId": null, "seconds": null, "promptVersion": null }, "componentId": "31254f08-e261-4b96-bbb6-c9cb2cad2f43", "parentComponentId": null }, { "bbox": { "h": 0.15, "w": 0.9, "x": 0.05, "y": 0.2 }, "kind": "paragraph", "text": "Accident: Designer had in mind a certain informal notion of objective/task to be performed, actual system failed at this catastrophically.", "attrs": null, "canon": { "kind": { "name": "Paragraph", "slug": "paragraph", "status": "active", "canonId": "5f2affc0-941d-4d07-959d-4bfc1930cea4", "version": 1, "description": "Single prose block in the body of a slide." }, "tool": null, "subkind": { "name": "Paragraph", "slug": "paragraph", "status": "active", "canonId": "019de52c-fb80-75af-aa5c-a7a4bfac2433", "version": 1, "description": null }, "framework": null }, "subkind": "paragraph", "groupRole": null, "confidence": null, "extraction": { "at": "2026-05-02 14:21:43.374+00", "model": "google/gemini-3.1-flash-lite-preview-20260303", "runId": null, "seconds": null, "promptVersion": null }, "componentId": "1b2ae06a-6a73-48be-a02a-cb1ff41aa858", "parentComponentId": null }, { "bbox": { "h": 0.1, "w": 0.4, "x": 0.05, "y": 0.05 }, "kind": "title", "text": "Sources of Accident Risk", "attrs": null, "canon": { "kind": { "name": "Title", "slug": "title", "status": "active", "canonId": "019de52c-f7dd-76d9-b848-b1f12bd85cf4", "version": 1, "description": "Slide title or headline." }, "tool": null, "subkind": { "name": "Headline", "slug": "headline", "status": "active", "canonId": "019de52c-fae1-7463-9a7c-d60668eced0d", "version": 1, "description": null }, "framework": null }, "subkind": "headline", "groupRole": null, "confidence": null, "extraction": { "at": "2026-05-02 14:21:43.374+00", "model": "google/gemini-3.1-flash-lite-preview-20260303", "runId": null, "seconds": null, "promptVersion": null }, "componentId": "1dc111e7-b013-45d0-bf8b-e753ce79ee1a", "parentComponentId": null } ], "metrics": [], "tools": [ { "tool": { "name": "Issue Tree", "slug": "issue-tree", "status": "active", "bestFor": null, "canonId": "019dd956-82e3-775d-a2b6-318154564b53", "version": 1, "bodyDocId": null, "description": "Breaking down a problem into component parts systematically", "familyLabel": null, "categoryName": "Block", "categorySlug": "block" }, "agent": "Architect", "layer": "slide", "agents": [ "Architect" ], "matchId": "6f9741ac-f9d3-4c1a-94cd-c13134720671", "evidence": "list/bullet: Causes: A. Wrong objective function B. Right objective function, problem in learning or inference C. (Non-ML problems -- e.g. software implementation bug)", "pageRefs": null, "priority": null, "whenToUse": null, "confidence": 0.7, "extraction": { "at": "2026-07-16 23:23:08.669463+00", "model": "or:meta-llama/llama-4-scout", "runId": null, "seconds": null, "promptVersion": null }, "pageNumber": 4, "whyItWorks": null, "antipattern": null, "cardinality": null, "narrativePurpose": null } ], "frameworks": [], "arcBeats": [ { "to": 4, "arc": { "name": "Problem-Agitate-Solution", "slug": "problem-agitate-solution", "status": "active", "bestFor": "Sales pitches, urgency-driven, competitive displacement", "canonId": "019dd956-5ec6-70f9-9fd8-fdbaaff39acf", "version": 1, "bodyDocId": "019df22a-2356-73eb-8061-d8c81dc0aa49", "structure": "Problem (Identify pain) -> Agitate (Make it worse) -> Solution (Provide relief)", "description": null, "familyLabel": null }, "from": 2, "name": "Problem", "beatId": "625ee8f7-4f93-455d-b525-f9a485b0c7c6", "matchId": "04c30632-8456-4dc0-8a7a-f7d40f1c761f", "beatType": { "name": "Problem (Identify pain)", "slug": "problem-agitate-solution-problem-identify-pain", "status": "active", "canonId": "019dd9b8-0e78-74bb-b8f6-54348367a131", "version": 1, "description": null }, "evidence": "The deck identifies potential negative social impacts of AI and sources of accident risk.", "position": 0, "isPrimaryArc": true, "parentBeatType": { "name": "Complication", "slug": "complication", "status": "active", "canonId": "019dd9b8-0275-7709-b2d6-4f77c3fc7579", "version": 1, "description": "The point in the narrative where the audience should feel discomfort. Required for argumentative arcs." }, "alignedBlockIds": null, "matchConfidence": 0.9 } ], "loops": [ { "to": 6, "from": 2, "loop": { "name": "27_cost_of_inaction", "slug": "27-cost-of-inaction", "status": "active", "bestFor": "Urgent budget requests, compliance, risk mitigation", "canonId": "019dd956-70d1-7395-a15b-857ba858b394", "version": 1, "bodyDocId": "019df22a-2420-77be-bc41-ded96d08cb21", "structure": "The Status Quo -> The Hidden Costs Accumulating -> The Future State of Inaction -> The Tipping Point", "description": "Quantify what happens if the audience does nothing", "familyLabel": null }, "matchId": "17544e50-3a29-4ebc-91be-39b72c52ee43", "evidence": "The deck highlights the potential negative social impacts of AI and the risks of not addressing them.", "position": 0, "objective": "What are the risks of not addressing AI safety?", "confidence": 0.8, "extraction": { "at": "2026-07-16 22:06:19.16729+00", "model": "or:meta-llama/llama-4-scout", "runId": null, "seconds": null, "promptVersion": "doc-narrative-v1" } } ], "locked": true }, { "page": 5, "type": "appendix", "title": "The slide shows a 'before' (left, dark/night mode) and 'after' (right, bright/day mode) of a racing game environment, likely used to test AI agent performance u", "function": "illustrate_case", "rawType": "other", "block": null, "metadata": { "slideType": "other", "slideTypeCanon": { "name": "Other (fallback)", "slug": "other", "status": "active", "canonId": "019de52d-0af5-71ed-83de-d51ae245fc43", "version": 1, "description": null }, "function": "illustrate_case", "functionCanon": { "name": "Illustrate case", "slug": "illustrate_case", "status": "active", "canonId": "019de52d-13b4-772c-bb73-1873612a7463", "version": 1, "description": null }, "density": "sparse", "densityScore": 15, "componentCount": 3, "textChars": 128, "nDataPoints": 0, "notes": "The slide shows a 'before' (left, dark/night mode) and 'after' (right, bright/day mode) of a racing game environment, likely used to test AI agent performance under varying visual conditions.", "elementsJson": null, "confidence": 0.9, "extraction": { "model": "google/gemini-3.1-flash-lite-preview-20260303", "runId": null, "at": "2026-05-02 14:21:40.306+00", "seconds": 2.31, "promptVersion": "v2-canon-conditioned" } }, "links": { "slide": "/slides/019de06f-84e4-70e4-bce4-3c7c4f09a8ea/5", "deck": "/decks/019de06f-84e4-70e4-bce4-3c7c4f09a8ea", "deckAnchor": "/decks/019de06f-84e4-70e4-bce4-3c7c4f09a8ea#slide-5", "deckJson": "/decks/019de06f-84e4-70e4-bce4-3c7c4f09a8ea.json" }, "components": [ { "bbox": { "h": 0.6, "w": 0.45, "x": 0.02, "y": 0.26 }, "kind": "image", "text": "Racing game interface in dark mode with performance charts", "attrs": null, "canon": { "kind": { "name": "Image", "slug": "image", "status": "active", "canonId": "019de52c-f978-701f-a720-51d029d769e7", "version": 1, "description": "Photo, illustration, screenshot or icon." }, "tool": null, "subkind": { "name": "Screenshot", "slug": "screenshot", "status": "active", "canonId": "019de52c-fd87-7509-8391-3d79fb339379", "version": 1, "description": null }, "framework": null }, "subkind": "screenshot", "groupRole": null, "confidence": null, "extraction": { "at": "2026-05-02 14:21:40.306+00", "model": "google/gemini-3.1-flash-lite-preview-20260303", "runId": null, "seconds": null, "promptVersion": null }, "componentId": "263e5c6b-64ae-4351-b2cf-25c40521eb54", "parentComponentId": null }, { "bbox": { "h": 0.6, "w": 0.45, "x": 0.51, "y": 0.26 }, "kind": "image", "text": "Racing game interface in bright mode", "attrs": null, "canon": { "kind": { "name": "Image", "slug": "image", "status": "active", "canonId": "019de52c-f978-701f-a720-51d029d769e7", "version": 1, "description": "Photo, illustration, screenshot or icon." }, "tool": null, "subkind": { "name": "Screenshot", "slug": "screenshot", "status": "active", "canonId": "019de52c-fd87-7509-8391-3d79fb339379", "version": 1, "description": null }, "framework": null }, "subkind": "screenshot", "groupRole": null, "confidence": null, "extraction": { "at": "2026-05-02 14:21:40.306+00", "model": "google/gemini-3.1-flash-lite-preview-20260303", "runId": null, "seconds": null, "promptVersion": null }, "componentId": "9ff6f52a-7e74-414b-9890-cfb65a986caa", "parentComponentId": null }, { "bbox": { "h": 0.05, "w": 0.4, "x": 0.04, "y": 0.08 }, "kind": "title", "text": "Robustness to distributional shift", "attrs": null, "canon": { "kind": { "name": "Title", "slug": "title", "status": "active", "canonId": "019de52c-f7dd-76d9-b848-b1f12bd85cf4", "version": 1, "description": "Slide title or headline." }, "tool": null, "subkind": { "name": "Headline", "slug": "headline", "status": "active", "canonId": "019de52c-fae1-7463-9a7c-d60668eced0d", "version": 1, "description": null }, "framework": null }, "subkind": "headline", "groupRole": null, "confidence": null, "extraction": { "at": "2026-05-02 14:21:40.306+00", "model": "google/gemini-3.1-flash-lite-preview-20260303", "runId": null, "seconds": null, "promptVersion": null }, "componentId": "6693ab49-d419-40c8-8f15-816f4bceb866", "parentComponentId": null } ], "metrics": [], "tools": [], "frameworks": [], "arcBeats": [ { "to": 6, "arc": { "name": "Problem-Agitate-Solution", "slug": "problem-agitate-solution", "status": "active", "bestFor": "Sales pitches, urgency-driven, competitive displacement", "canonId": "019dd956-5ec6-70f9-9fd8-fdbaaff39acf", "version": 1, "bodyDocId": "019df22a-2356-73eb-8061-d8c81dc0aa49", "structure": "Problem (Identify pain) -> Agitate (Make it worse) -> Solution (Provide relief)", "description": null, "familyLabel": null }, "from": 5, "name": "Agitate", "beatId": "0c5776c9-81ed-47ea-a4d1-51709ebdb4ed", "matchId": "04c30632-8456-4dc0-8a7a-f7d40f1c761f", "beatType": { "name": "Agitate (Make it worse)", "slug": "problem-agitate-solution-agitate-make-it-worse", "status": "active", "canonId": "019dd9b8-0f5a-71dd-bae7-8fadb0d09456", "version": 1, "description": null }, "evidence": "The deck highlights the risks of distributional shift and bad objective functions.", "position": 1, "isPrimaryArc": true, "parentBeatType": { "name": "Development", "slug": "development", "status": "active", "canonId": "019dd9b8-0332-7008-b3aa-11b0840a57b1", "version": 1, "description": "The middle stretch in arcs that have one. Often where length problems creep in." }, "alignedBlockIds": null, "matchConfidence": 0.9 } ], "loops": [ { "to": 6, "from": 2, "loop": { "name": "27_cost_of_inaction", "slug": "27-cost-of-inaction", "status": "active", "bestFor": "Urgent budget requests, compliance, risk mitigation", "canonId": "019dd956-70d1-7395-a15b-857ba858b394", "version": 1, "bodyDocId": "019df22a-2420-77be-bc41-ded96d08cb21", "structure": "The Status Quo -> The Hidden Costs Accumulating -> The Future State of Inaction -> The Tipping Point", "description": "Quantify what happens if the audience does nothing", "familyLabel": null }, "matchId": "17544e50-3a29-4ebc-91be-39b72c52ee43", "evidence": "The deck highlights the potential negative social impacts of AI and the risks of not addressing them.", "position": 0, "objective": "What are the risks of not addressing AI safety?", "confidence": 0.8, "extraction": { "at": "2026-07-16 22:06:19.16729+00", "model": "or:meta-llama/llama-4-scout", "runId": null, "seconds": null, "promptVersion": "doc-narrative-v1" } } ], "locked": true }, { "page": 6, "type": "tension", "title": "The slide uses a screenshot of a CNET article to provide concrete evidence of a failure in AI model performance.", "function": "diagnose_problem", "rawType": "problem_statement", "block": null, "metadata": { "slideType": "problem_statement", "slideTypeCanon": { "name": "Problem statement", "slug": "problem_statement", "status": "active", "canonId": "019de52d-0879-768d-bba8-8da8a5a121d4", "version": 1, "description": null }, "function": "diagnose_problem", "functionCanon": { "name": "Diagnose problem", "slug": "diagnose_problem", "status": "active", "canonId": "019de52d-176d-761f-b608-cf52262d2ab2", "version": 1, "description": null }, "density": "sparse", "densityScore": 11, "componentCount": 2, "textChars": 139, "nDataPoints": 0, "notes": "The slide uses a screenshot of a CNET article to provide concrete evidence of a failure in AI model performance.", "elementsJson": null, "confidence": 1, "extraction": { "model": "google/gemini-3.1-flash-lite-preview-20260303", "runId": null, "at": "2026-05-02 14:21:40.046+00", "seconds": 1.886, "promptVersion": "v2-canon-conditioned" } }, "links": { "slide": "/slides/019de06f-84e4-70e4-bce4-3c7c4f09a8ea/6", "deck": "/decks/019de06f-84e4-70e4-bce4-3c7c4f09a8ea", "deckAnchor": "/decks/019de06f-84e4-70e4-bce4-3c7c4f09a8ea#slide-6", "deckJson": "/decks/019de06f-84e4-70e4-bce4-3c7c4f09a8ea.json" }, "components": [ { "bbox": { "h": 0.75, "w": 0.8, "x": 0.1, "y": 0.18 }, "kind": "image", "text": "CNET article screenshot: Google apologizes for algorithm mistakenly calling black people 'gorillas'", "attrs": null, "canon": { "kind": { "name": "Image", "slug": "image", "status": "active", "canonId": "019de52c-f978-701f-a720-51d029d769e7", "version": 1, "description": "Photo, illustration, screenshot or icon." }, "tool": null, "subkind": { "name": "Screenshot", "slug": "screenshot", "status": "active", "canonId": "019de52c-fd87-7509-8391-3d79fb339379", "version": 1, "description": null }, "framework": null }, "subkind": "screenshot", "groupRole": null, "confidence": null, "extraction": { "at": "2026-05-02 14:21:40.046+00", "model": "google/gemini-3.1-flash-lite-preview-20260303", "runId": null, "seconds": null, "promptVersion": null }, "componentId": "854b18ca-9cd2-4c85-b709-4daa136f98f1", "parentComponentId": null }, { "bbox": { "h": 0.08, "w": 0.5, "x": 0.05, "y": 0.05 }, "kind": "title", "text": "Bad Consequences of Distributional Shift", "attrs": null, "canon": { "kind": { "name": "Title", "slug": "title", "status": "active", "canonId": "019de52c-f7dd-76d9-b848-b1f12bd85cf4", "version": 1, "description": "Slide title or headline." }, "tool": null, "subkind": { "name": "Headline", "slug": "headline", "status": "active", "canonId": "019de52c-fae1-7463-9a7c-d60668eced0d", "version": 1, "description": null }, "framework": null }, "subkind": "headline", "groupRole": null, "confidence": null, "extraction": { "at": "2026-05-02 14:21:40.046+00", "model": "google/gemini-3.1-flash-lite-preview-20260303", "runId": null, "seconds": null, "promptVersion": null }, "componentId": "0fccecd3-0d70-441b-9c14-38f8236c742b", "parentComponentId": null } ], "metrics": [], "tools": [ { "tool": { "name": "Before-After-Bridge", "slug": "before-after-bridge", "status": "active", "bestFor": null, "canonId": "019dd956-8604-7188-883f-42370288ad05", "version": 1, "bodyDocId": null, "description": "Narrative structure: Paint the before, show the after, explain the bridge", "familyLabel": null, "categoryName": "Block", "categorySlug": "block" }, "agent": "Storyteller", "layer": "slide", "agents": [ "Storyteller" ], "matchId": "8b7db23b-7b58-4c89-b70d-814fe4487ec1", "evidence": "image/screenshot: CNET article screenshot: Google apologizes for algorithm mistakenly calling black people 'gorillas'", "pageRefs": null, "priority": null, "whenToUse": null, "confidence": 0.7, "extraction": { "at": "2026-07-16 23:23:08.782101+00", "model": "or:meta-llama/llama-4-scout", "runId": null, "seconds": null, "promptVersion": null }, "pageNumber": 6, "whyItWorks": null, "antipattern": null, "cardinality": null, "narrativePurpose": null } ], "frameworks": [], "arcBeats": [ { "to": 6, "arc": { "name": "Problem-Agitate-Solution", "slug": "problem-agitate-solution", "status": "active", "bestFor": "Sales pitches, urgency-driven, competitive displacement", "canonId": "019dd956-5ec6-70f9-9fd8-fdbaaff39acf", "version": 1, "bodyDocId": "019df22a-2356-73eb-8061-d8c81dc0aa49", "structure": "Problem (Identify pain) -> Agitate (Make it worse) -> Solution (Provide relief)", "description": null, "familyLabel": null }, "from": 5, "name": "Agitate", "beatId": "0c5776c9-81ed-47ea-a4d1-51709ebdb4ed", "matchId": "04c30632-8456-4dc0-8a7a-f7d40f1c761f", "beatType": { "name": "Agitate (Make it worse)", "slug": "problem-agitate-solution-agitate-make-it-worse", "status": "active", "canonId": "019dd9b8-0f5a-71dd-bae7-8fadb0d09456", "version": 1, "description": null }, "evidence": "The deck highlights the risks of distributional shift and bad objective functions.", "position": 1, "isPrimaryArc": true, "parentBeatType": { "name": "Development", "slug": "development", "status": "active", "canonId": "019dd9b8-0332-7008-b3aa-11b0840a57b1", "version": 1, "description": "The middle stretch in arcs that have one. Often where length problems creep in." }, "alignedBlockIds": null, "matchConfidence": 0.9 } ], "loops": [ { "to": 6, "from": 2, "loop": { "name": "27_cost_of_inaction", "slug": "27-cost-of-inaction", "status": "active", "bestFor": "Urgent budget requests, compliance, risk mitigation", "canonId": "019dd956-70d1-7395-a15b-857ba858b394", "version": 1, "bodyDocId": "019df22a-2420-77be-bc41-ded96d08cb21", "structure": "The Status Quo -> The Hidden Costs Accumulating -> The Future State of Inaction -> The Tipping Point", "description": "Quantify what happens if the audience does nothing", "familyLabel": null }, "matchId": "17544e50-3a29-4ebc-91be-39b72c52ee43", "evidence": "The deck highlights the potential negative social impacts of AI and the risks of not addressing them.", "position": 0, "objective": "What are the risks of not addressing AI safety?", "confidence": 0.8, "extraction": { "at": "2026-07-16 22:06:19.16729+00", "model": "or:meta-llama/llama-4-scout", "runId": null, "seconds": null, "promptVersion": "doc-narrative-v1" } } ], "locked": true }, { "page": 7, "type": "tension", "title": "The slide explicitly references Goodhart's Law as a theoretical underpinning for the problem.", "function": "frame_problem", "rawType": "problem_statement", "block": null, "metadata": { "slideType": "problem_statement", "slideTypeCanon": { "name": "Problem statement", "slug": "problem_statement", "status": "active", "canonId": "019de52d-0879-768d-bba8-8da8a5a121d4", "version": 1, "description": null }, "function": "frame_problem", "functionCanon": { "name": "Frame problem", "slug": "frame_problem", "status": "active", "canonId": "019de52d-13db-753e-ad70-e9b2d0dab398", "version": 1, "description": null }, "density": "dense", "densityScore": 66, "componentCount": 6, "textChars": 564, "nDataPoints": 0, "notes": "The slide explicitly references Goodhart's Law as a theoretical underpinning for the problem.", "elementsJson": null, "confidence": 1, "extraction": { "model": "google/gemini-3.1-flash-lite-preview-20260303", "runId": null, "at": "2026-05-02 14:21:43.605+00", "seconds": 5.175, "promptVersion": "v2-canon-conditioned" } }, "links": { "slide": "/slides/019de06f-84e4-70e4-bce4-3c7c4f09a8ea/7", "deck": "/decks/019de06f-84e4-70e4-bce4-3c7c4f09a8ea", "deckAnchor": "/decks/019de06f-84e4-70e4-bce4-3c7c4f09a8ea#slide-7", "deckJson": "/decks/019de06f-84e4-70e4-bce4-3c7c4f09a8ea.json" }, "components": [ { "bbox": { "h": 0.1, "w": 0.9, "x": 0.05, "y": 0.68 }, "kind": "list", "text": "We currently address this by finding simple proxies for these complex goals, but this can lead to unexpected and in fact very unsafe behavior.", "attrs": null, "canon": { "kind": { "name": "List", "slug": "list", "status": "active", "canonId": "019de52c-f94b-779b-a53b-fd4904a5f4d2", "version": 1, "description": "Bullet or numbered list." }, "tool": null, "subkind": { "name": "Bullet list", "slug": "bullet", "status": "active", "canonId": "019de52c-fc98-7169-a083-5cab805076ca", "version": 1, "description": null }, "framework": null }, "subkind": "bullet", "groupRole": null, "confidence": null, "extraction": { "at": "2026-05-02 14:21:43.605+00", "model": "google/gemini-3.1-flash-lite-preview-20260303", "runId": null, "seconds": null, "promptVersion": null }, "componentId": "06599305-1004-4731-a0a5-389733bc1c93", "parentComponentId": null }, { "bbox": { "h": 0.1, "w": 0.85, "x": 0.05, "y": 0.2 }, "kind": "list", "text": "Our current AI systems have a curious limitation: behavior can be very complex, but goal has to be simple: consider Chess, Go.", "attrs": null, "canon": { "kind": { "name": "List", "slug": "list", "status": "active", "canonId": "019de52c-f94b-779b-a53b-fd4904a5f4d2", "version": 1, "description": "Bullet or numbered list." }, "tool": null, "subkind": { "name": "Bullet list", "slug": "bullet", "status": "active", "canonId": "019de52c-fc98-7169-a083-5cab805076ca", "version": 1, "description": null }, "framework": null }, "subkind": "bullet", "groupRole": null, "confidence": null, "extraction": { "at": "2026-05-02 14:21:43.605+00", "model": "google/gemini-3.1-flash-lite-preview-20260303", "runId": null, "seconds": null, "promptVersion": null }, "componentId": "0a75775c-8b2c-4007-a87b-b700feccc75d", "parentComponentId": null }, { "bbox": { "h": 0.05, "w": 0.6, "x": 0.05, "y": 0.35 }, "kind": "list", "text": "Real world tasks often have high-level, fuzzy, complicated goals", "attrs": null, "canon": { "kind": { "name": "List", "slug": "list", "status": "active", "canonId": "019de52c-f94b-779b-a53b-fd4904a5f4d2", "version": 1, "description": "Bullet or numbered list." }, "tool": null, "subkind": { "name": "Bullet list", "slug": "bullet", "status": "active", "canonId": "019de52c-fc98-7169-a083-5cab805076ca", "version": 1, "description": null }, "framework": null }, "subkind": "bullet", "groupRole": null, "confidence": null, "extraction": { "at": "2026-05-02 14:21:43.605+00", "model": "google/gemini-3.1-flash-lite-preview-20260303", "runId": null, "seconds": null, "promptVersion": null }, "componentId": "0c412fbb-cb6c-41df-8741-20e37375304c", "parentComponentId": null }, { "bbox": { "h": 0.05, "w": 0.8, "x": 0.05, "y": 0.85 }, "kind": "list", "text": "Goodhart's law: \"When a metric becomes a target, it ceases to be a good", "attrs": null, "canon": { "kind": { "name": "List", "slug": "list", "status": "active", "canonId": "019de52c-f94b-779b-a53b-fd4904a5f4d2", "version": 1, "description": "Bullet or numbered list." }, "tool": null, "subkind": { "name": "Bullet list", "slug": "bullet", "status": "active", "canonId": "019de52c-fc98-7169-a083-5cab805076ca", "version": 1, "description": null }, "framework": null }, "subkind": "bullet", "groupRole": null, "confidence": null, "extraction": { "at": "2026-05-02 14:21:43.605+00", "model": "google/gemini-3.1-flash-lite-preview-20260303", "runId": null, "seconds": null, "promptVersion": null }, "componentId": "11c388ef-e1c1-47bd-8a7f-950ac37484c9", "parentComponentId": null }, { "bbox": { "h": 0.2, "w": 0.5, "x": 0.1, "y": 0.42 }, "kind": "list", "text": "Arrange the furniture in a room to look nice; Engage in realistic and informative dialogue with humans; Plan and execute a mission to Mars", "attrs": null, "canon": { "kind": { "name": "List", "slug": "list", "status": "active", "canonId": "019de52c-f94b-779b-a53b-fd4904a5f4d2", "version": 1, "description": "Bullet or numbered list." }, "tool": null, "subkind": { "name": "Bullet list", "slug": "bullet", "status": "active", "canonId": "019de52c-fc98-7169-a083-5cab805076ca", "version": 1, "description": null }, "framework": null }, "subkind": "bullet", "groupRole": null, "confidence": null, "extraction": { "at": "2026-05-02 14:21:43.605+00", "model": "google/gemini-3.1-flash-lite-preview-20260303", "runId": null, "seconds": null, "promptVersion": null }, "componentId": "5045bc3b-17ec-47a0-86fa-aa87d92beabf", "parentComponentId": null }, { "bbox": { "h": 0.06, "w": 0.4, "x": 0.05, "y": 0.08 }, "kind": "title", "text": "Bad Objective Functions", "attrs": null, "canon": { "kind": { "name": "Title", "slug": "title", "status": "active", "canonId": "019de52c-f7dd-76d9-b848-b1f12bd85cf4", "version": 1, "description": "Slide title or headline." }, "tool": null, "subkind": { "name": "Headline", "slug": "headline", "status": "active", "canonId": "019de52c-fae1-7463-9a7c-d60668eced0d", "version": 1, "description": null }, "framework": null }, "subkind": "headline", "groupRole": null, "confidence": null, "extraction": { "at": "2026-05-02 14:21:43.605+00", "model": "google/gemini-3.1-flash-lite-preview-20260303", "runId": null, "seconds": null, "promptVersion": null }, "componentId": "bc7b0728-3431-485d-82b8-58f80f5cc1d6", "parentComponentId": null } ], "metrics": [], "tools": [ { "tool": { "name": "Goodhart's Law", "slug": "goodharts-law", "status": "active", "bestFor": null, "canonId": "019dd956-8ed1-7421-ab8c-49dbce192ef5", "version": 1, "bodyDocId": null, "description": "When a measure becomes a target it ceases to be a good measure", "familyLabel": null, "categoryName": "Block", "categorySlug": "block" }, "agent": "Architect", "layer": "slide", "agents": [ "Architect" ], "matchId": "1ce75a51-24f8-4120-add4-501fd9568a00", "evidence": "Goodhart's law: \"When a metric becomes a target, it ceases to be a good", "pageRefs": null, "priority": null, "whenToUse": null, "confidence": 0.9, "extraction": { "at": "2026-07-16 23:23:08.896334+00", "model": "or:meta-llama/llama-4-scout", "runId": null, "seconds": null, "promptVersion": null }, "pageNumber": 7, "whyItWorks": null, "antipattern": null, "cardinality": null, "narrativePurpose": null }, { "tool": { "name": "List presentation", "slug": "list-presentation", "status": "active", "bestFor": null, "canonId": "019dd9e1-4ef7-777d-ba68-9074c2b3bcf1", "version": 1, "bodyDocId": null, "description": "Bulleted, numbered, or checklist enumerations.", "familyLabel": null, "categoryName": null, "categorySlug": null }, "agent": null, "layer": "slide", "agents": null, "matchId": "8292e732-8e47-41f6-91c9-c3812e4d2ad4", "evidence": "list/bullet: We currently address this by finding simple proxies for these complex goals, but this can lead to unexpected and in fact very unsafe behavior.", "pageRefs": null, "priority": null, "whenToUse": null, "confidence": 0.8, "extraction": { "at": "2026-07-16 23:23:08.857469+00", "model": "or:meta-llama/llama-4-scout", "runId": null, "seconds": null, "promptVersion": null }, "pageNumber": 7, "whyItWorks": null, "antipattern": null, "cardinality": "common", "narrativePurpose": null } ], "frameworks": [ { "matchId": "0c07c704-2b3b-4527-9309-319ee0362e41", "evidence": "Explicitly cited in the final bullet point.", "framework": null, "confidence": 1, "extraction": { "at": "2026-05-02 14:21:43.605+00", "model": "google/gemini-3.1-flash-lite-preview-20260303", "runId": null, "seconds": null, "promptVersion": null }, "pageNumber": 7, "frameworkName": "goodharts-law" } ], "arcBeats": [], "loops": [], "locked": true }, { "page": 8, "type": "setup", "title": "The image is a screenshot of a boat racing game.", "function": "propose_solution", "rawType": "propose_solution", "block": null, "metadata": { "slideType": "propose_solution", "slideTypeCanon": { "name": "Propose solution (contrarian)", "slug": "propose_solution", "status": "active", "canonId": "019de52d-1274-703c-81c6-5550f451697b", "version": 1, "description": null }, "function": "propose_solution", "functionCanon": { "name": "Propose solution (contrarian)", "slug": "propose_solution", "status": "active", "canonId": "019de52d-185b-74e9-bdfd-ea8cf318563f", "version": 1, "description": null }, "density": "sparse", "densityScore": 13, "componentCount": 3, "textChars": 97, "nDataPoints": 0, "notes": "The image is a screenshot of a boat racing game.", "elementsJson": null, "confidence": 0.9, "extraction": { "model": "google/gemini-3.1-flash-lite-preview-20260303", "runId": null, "at": "2026-05-02 14:21:42.724+00", "seconds": 4.238, "promptVersion": "v2-canon-conditioned" } }, "links": { "slide": "/slides/019de06f-84e4-70e4-bce4-3c7c4f09a8ea/8", "deck": "/decks/019de06f-84e4-70e4-bce4-3c7c4f09a8ea", "deckAnchor": "/decks/019de06f-84e4-70e4-bce4-3c7c4f09a8ea#slide-8", "deckJson": "/decks/019de06f-84e4-70e4-bce4-3c7c4f09a8ea.json" }, "components": [ { "bbox": { "h": 0.65, "w": 0.56, "x": 0.22, "y": 0.15 }, "kind": "image", "text": "Video game gameplay screenshot", "attrs": null, "canon": { "kind": { "name": "Image", "slug": "image", "status": "active", "canonId": "019de52c-f978-701f-a720-51d029d769e7", "version": 1, "description": "Photo, illustration, screenshot or icon." }, "tool": null, "subkind": { "name": "Screenshot", "slug": "screenshot", "status": "active", "canonId": "019de52c-fd87-7509-8391-3d79fb339379", "version": 1, "description": null }, "framework": null }, "subkind": "screenshot", "groupRole": null, "confidence": null, "extraction": { "at": "2026-05-02 14:21:42.724+00", "model": "google/gemini-3.1-flash-lite-preview-20260303", "runId": null, "seconds": null, "promptVersion": null }, "componentId": "bda386b4-9dd0-4c2f-9160-e00ce58063c6", "parentComponentId": null }, { "bbox": { "h": 0.05, "w": 0.3, "x": 0.05, "y": 0.85 }, "kind": "paragraph", "text": "(this is a human playing)", "attrs": null, "canon": { "kind": { "name": "Paragraph", "slug": "paragraph", "status": "active", "canonId": "5f2affc0-941d-4d07-959d-4bfc1930cea4", "version": 1, "description": "Single prose block in the body of a slide." }, "tool": null, "subkind": { "name": "Paragraph", "slug": "paragraph", "status": "active", "canonId": "019de52c-fb80-75af-aa5c-a7a4bfac2433", "version": 1, "description": null }, "framework": null }, "subkind": "paragraph", "groupRole": null, "confidence": null, "extraction": { "at": "2026-05-02 14:21:42.724+00", "model": "google/gemini-3.1-flash-lite-preview-20260303", "runId": null, "seconds": null, "promptVersion": null }, "componentId": "6ca0d146-3b54-4782-9247-93886a2bc623", "parentComponentId": null }, { "bbox": { "h": 0.1, "w": 0.5, "x": 0.05, "y": 0.05 }, "kind": "title", "text": "We'd like to train an ML system to do this", "attrs": null, "canon": { "kind": { "name": "Title", "slug": "title", "status": "active", "canonId": "019de52c-f7dd-76d9-b848-b1f12bd85cf4", "version": 1, "description": "Slide title or headline." }, "tool": null, "subkind": { "name": "Headline", "slug": "headline", "status": "active", "canonId": "019de52c-fae1-7463-9a7c-d60668eced0d", "version": 1, "description": null }, "framework": null }, "subkind": "headline", "groupRole": null, "confidence": null, "extraction": { "at": "2026-05-02 14:21:42.724+00", "model": "google/gemini-3.1-flash-lite-preview-20260303", "runId": null, "seconds": null, "promptVersion": null }, "componentId": "ff83c059-5851-4018-87a2-95b17c40a7cb", "parentComponentId": null } ], "metrics": [], "tools": [], "frameworks": [], "arcBeats": [ { "to": 11, "arc": { "name": "Problem-Agitate-Solution", "slug": "problem-agitate-solution", "status": "active", "bestFor": "Sales pitches, urgency-driven, competitive displacement", "canonId": "019dd956-5ec6-70f9-9fd8-fdbaaff39acf", "version": 1, "bodyDocId": "019df22a-2356-73eb-8061-d8c81dc0aa49", "structure": "Problem (Identify pain) -> Agitate (Make it worse) -> Solution (Provide relief)", "description": null, "familyLabel": null }, "from": 8, "name": "Solution", "beatId": "72a1fd78-bc62-4d21-a603-262dd269d030", "matchId": "04c30632-8456-4dc0-8a7a-f7d40f1c761f", "beatType": { "name": "Solution (Provide relief)", "slug": "problem-agitate-solution-solution-provide-relief", "status": "active", "canonId": "019dd9b8-0fee-73e9-9ae5-799c7f3895cd", "version": 1, "description": null }, "evidence": "The deck proposes a solution using human feedback training processes and provides evidence to support it.", "position": 2, "isPrimaryArc": true, "parentBeatType": { "name": "Resolution", "slug": "resolution", "status": "active", "canonId": "019dd9b8-045d-703b-8dd7-e92aba3ba91b", "version": 1, "description": "How the deck ends. Required in 19/20 arcs. The exception (Sparkline) ends with new bliss which is functionally a resolution." }, "alignedBlockIds": null, "matchConfidence": 0.9 } ], "loops": [ { "to": 14, "from": 8, "loop": { "name": "53_jobs_to_be_done", "slug": "53-jobs-to-be-done", "status": "active", "bestFor": "Product innovation, market entry, competitive positioning", "canonId": "019dd956-7b6c-7594-b5b3-eecbe0ab48e8", "version": 1, "bodyDocId": "019df22a-2420-77be-bc41-ded96d08cb21", "structure": "The Customer's Job -> Current Solutions (Hired/Fired) -> Unmet Needs -> Our Solution Fit", "description": "Reframe the problem around what the customer is trying to accomplish, not what they're buying", "familyLabel": null }, "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?", "confidence": 0.7, "extraction": { "at": "2026-07-16 22:06:19.201225+00", "model": "or:meta-llama/llama-4-scout", "runId": null, "seconds": null, "promptVersion": "doc-narrative-v1" } } ], "locked": true }, { "page": 9, "type": "evidence", "title": "The slide features a visual representation of a simulated agent (likely a MuJoCo environment) used in the referenced research.", "function": "cite_precedent", "rawType": "cite_precedent", "block": null, "metadata": { "slideType": "cite_precedent", "slideTypeCanon": { "name": "Cite precedent", "slug": "cite_precedent", "status": "active", "canonId": "019de52d-12ec-72be-8881-51b438fa76c7", "version": 1, "description": null }, "function": "cite_precedent", "functionCanon": { "name": "Cite precedent", "slug": "cite_precedent", "status": "active", "canonId": "019de52d-142b-714b-86d4-58aad7b568ec", "version": 1, "description": null }, "density": "sparse", "densityScore": 11, "componentCount": 2, "textChars": 147, "nDataPoints": 0, "notes": "The slide features a visual representation of a simulated agent (likely a MuJoCo environment) used in the referenced research.", "elementsJson": null, "confidence": 1, "extraction": { "model": "google/gemini-3.1-flash-lite-preview-20260303", "runId": null, "at": "2026-05-02 14:21:41.083+00", "seconds": 2.415, "promptVersion": "v2-canon-conditioned" } }, "links": { "slide": "/slides/019de06f-84e4-70e4-bce4-3c7c4f09a8ea/9", "deck": "/decks/019de06f-84e4-70e4-bce4-3c7c4f09a8ea", "deckAnchor": "/decks/019de06f-84e4-70e4-bce4-3c7c4f09a8ea#slide-9", "deckJson": "/decks/019de06f-84e4-70e4-bce4-3c7c4f09a8ea.json" }, "components": [ { "bbox": { "h": 0.8, "w": 0.6, "x": 0.2, "y": 0.05 }, "kind": "image", "text": null, "attrs": null, "canon": { "kind": { "name": "Image", "slug": "image", "status": "active", "canonId": "019de52c-f978-701f-a720-51d029d769e7", "version": 1, "description": "Photo, illustration, screenshot or icon." }, "tool": null, "subkind": { "name": "Illustration", "slug": "illustration", "status": "active", "canonId": "019de52c-fe76-7772-a5ac-c0b7294d7615", "version": 1, "description": null }, "framework": null }, "subkind": "illustration", "groupRole": null, "confidence": null, "extraction": { "at": "2026-05-02 14:21:41.083+00", "model": "google/gemini-3.1-flash-lite-preview-20260303", "runId": null, "seconds": null, "promptVersion": null }, "componentId": "70d7895c-77ee-4f98-b9d9-d367455e4bbc", "parentComponentId": null }, { "bbox": { "h": 0.08, "w": 0.9, "x": 0.05, "y": 0.9 }, "kind": "source-note", "text": "Paul Christiano, Jan Leike, Tom Brown, Miljan Martik, Shane Legg, Dario Amodei\nDeep Reinforcement Learning from Human Preferences, arXiv:1706.03741", "attrs": null, "canon": { "kind": { "name": "Source note", "slug": "source-note", "status": "active", "canonId": "23484952-494b-4a5e-b1ad-550d0d70e948", "version": 1, "description": "Attribution / source / caveat placed at the slide footer. Numbered footnotes use attrs.numbered." }, "tool": null, "subkind": null, "framework": null }, "subkind": null, "groupRole": null, "confidence": null, "extraction": { "at": "2026-05-02 14:21:41.083+00", "model": "google/gemini-3.1-flash-lite-preview-20260303", "runId": null, "seconds": null, "promptVersion": null }, "componentId": "321ae3f3-0d7c-4ad7-b417-8ddb74cdad3a", "parentComponentId": null } ], "metrics": [], "tools": [], "frameworks": [ { "matchId": "5c3de249-d9fa-4e70-9b26-356d456b6e21", "evidence": "Citing a seminal academic paper to establish credibility.", "framework": null, "confidence": 0.9, "extraction": { "at": "2026-05-02 14:21:41.083+00", "model": "google/gemini-3.1-flash-lite-preview-20260303", "runId": null, "seconds": null, "promptVersion": null }, "pageNumber": 9, "frameworkName": "authority-citation" } ], "arcBeats": [ { "to": 11, "arc": { "name": "Problem-Agitate-Solution", "slug": "problem-agitate-solution", "status": "active", "bestFor": "Sales pitches, urgency-driven, competitive displacement", "canonId": "019dd956-5ec6-70f9-9fd8-fdbaaff39acf", "version": 1, "bodyDocId": "019df22a-2356-73eb-8061-d8c81dc0aa49", "structure": "Problem (Identify pain) -> Agitate (Make it worse) -> Solution (Provide relief)", "description": null, "familyLabel": null }, "from": 8, "name": "Solution", "beatId": "72a1fd78-bc62-4d21-a603-262dd269d030", "matchId": "04c30632-8456-4dc0-8a7a-f7d40f1c761f", "beatType": { "name": "Solution (Provide relief)", "slug": "problem-agitate-solution-solution-provide-relief", "status": "active", "canonId": "019dd9b8-0fee-73e9-9ae5-799c7f3895cd", "version": 1, "description": null }, "evidence": "The deck proposes a solution using human feedback training processes and provides evidence to support it.", "position": 2, "isPrimaryArc": true, "parentBeatType": { "name": "Resolution", "slug": "resolution", "status": "active", "canonId": "019dd9b8-045d-703b-8dd7-e92aba3ba91b", "version": 1, "description": "How the deck ends. Required in 19/20 arcs. The exception (Sparkline) ends with new bliss which is functionally a resolution." }, "alignedBlockIds": null, "matchConfidence": 0.9 } ], "loops": [ { "to": 14, "from": 8, "loop": { "name": "53_jobs_to_be_done", "slug": "53-jobs-to-be-done", "status": "active", "bestFor": "Product innovation, market entry, competitive positioning", "canonId": "019dd956-7b6c-7594-b5b3-eecbe0ab48e8", "version": 1, "bodyDocId": "019df22a-2420-77be-bc41-ded96d08cb21", "structure": "The Customer's Job -> Current Solutions (Hired/Fired) -> Unmet Needs -> Our Solution Fit", "description": "Reframe the problem around what the customer is trying to accomplish, not what they're buying", "familyLabel": null }, "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?", "confidence": 0.7, "extraction": { "at": "2026-07-16 22:06:19.201225+00", "model": "or:meta-llama/llama-4-scout", "runId": null, "seconds": null, "promptVersion": "doc-narrative-v1" } } ], "locked": true }, { "page": 10, "type": "appendix", "title": "The slide shows a binary choice interface used for training AI models, specifically for reinforcement learning.", "function": "present_framework", "rawType": "other", "block": null, "metadata": { "slideType": "other", "slideTypeCanon": { "name": "Other (fallback)", "slug": "other", "status": "active", "canonId": "019de52d-0af5-71ed-83de-d51ae245fc43", "version": 1, "description": null }, "function": "present_framework", "functionCanon": { "name": "Present framework", "slug": "present_framework", "status": "active", "canonId": "019de52d-1403-7667-9100-d2ac6709f750", "version": 1, "description": null }, "density": "sparse", "densityScore": 19, "componentCount": 3, "textChars": 182, "nDataPoints": 0, "notes": "The slide shows a binary choice interface used for training AI models, specifically for reinforcement learning.", "elementsJson": null, "confidence": 0.9, "extraction": { "model": "google/gemini-3.1-flash-lite-preview-20260303", "runId": null, "at": "2026-05-02 14:21:43.585+00", "seconds": 4.909, "promptVersion": "v2-canon-conditioned" } }, "links": { "slide": "/slides/019de06f-84e4-70e4-bce4-3c7c4f09a8ea/10", "deck": "/decks/019de06f-84e4-70e4-bce4-3c7c4f09a8ea", "deckAnchor": "/decks/019de06f-84e4-70e4-bce4-3c7c4f09a8ea#slide-10", "deckJson": "/decks/019de06f-84e4-70e4-bce4-3c7c4f09a8ea.json" }, "components": [ { "bbox": { "h": 0.7, "w": 0.68, "x": 0.16, "y": 0.15 }, "kind": "image", "text": "Two side-by-side simulation windows with 'Left is better' and 'Right is better' buttons above them.", "attrs": null, "canon": { "kind": { "name": "Image", "slug": "image", "status": "active", "canonId": "019de52c-f978-701f-a720-51d029d769e7", "version": 1, "description": "Photo, illustration, screenshot or icon." }, "tool": null, "subkind": { "name": "Screenshot", "slug": "screenshot", "status": "active", "canonId": "019de52c-fd87-7509-8391-3d79fb339379", "version": 1, "description": null }, "framework": null }, "subkind": "screenshot", "groupRole": null, "confidence": null, "extraction": { "at": "2026-05-02 14:21:43.585+00", "model": "google/gemini-3.1-flash-lite-preview-20260303", "runId": null, "seconds": null, "promptVersion": null }, "componentId": "08aec86e-fe4b-42eb-8ed4-3294fc144f33", "parentComponentId": null }, { "bbox": { "h": 0.05, "w": 0.6, "x": 0.1, "y": 0.88 }, "kind": "paragraph", "text": "Requires only a few hundred binary bits of feedback!", "attrs": null, "canon": { "kind": { "name": "Paragraph", "slug": "paragraph", "status": "active", "canonId": "5f2affc0-941d-4d07-959d-4bfc1930cea4", "version": 1, "description": "Single prose block in the body of a slide." }, "tool": null, "subkind": { "name": "Paragraph", "slug": "paragraph", "status": "active", "canonId": "019de52c-fb80-75af-aa5c-a7a4bfac2433", "version": 1, "description": null }, "framework": null }, "subkind": "paragraph", "groupRole": null, "confidence": null, "extraction": { "at": "2026-05-02 14:21:43.585+00", "model": "google/gemini-3.1-flash-lite-preview-20260303", "runId": null, "seconds": null, "promptVersion": null }, "componentId": "8e78c0eb-5da5-430b-8531-6b79d7dcff89", "parentComponentId": null }, { "bbox": { "h": 0.1, "w": 0.4, "x": 0.05, "y": 0.05 }, "kind": "title", "text": "Human Feedback Training Process", "attrs": null, "canon": { "kind": { "name": "Title", "slug": "title", "status": "active", "canonId": "019de52c-f7dd-76d9-b848-b1f12bd85cf4", "version": 1, "description": "Slide title or headline." }, "tool": null, "subkind": { "name": "Headline", "slug": "headline", "status": "active", "canonId": "019de52c-fae1-7463-9a7c-d60668eced0d", "version": 1, "description": null }, "framework": null }, "subkind": "headline", "groupRole": null, "confidence": null, "extraction": { "at": "2026-05-02 14:21:43.585+00", "model": "google/gemini-3.1-flash-lite-preview-20260303", "runId": null, "seconds": null, "promptVersion": null }, "componentId": "5e985509-e8b0-4410-aa65-82e9fc9676bf", "parentComponentId": null } ], "metrics": [], "tools": [ { "tool": { "name": "Process flow", "slug": "process-flow", "status": "active", "bestFor": null, "canonId": "019dd9e1-4dca-76bc-8135-401836a1c4c0", "version": 1, "bodyDocId": null, "description": "Workflow, steps, sequential diagram.", "familyLabel": null, "categoryName": null, "categorySlug": null }, "agent": null, "layer": "slide", "agents": null, "matchId": "9f7c2a90-5eab-477b-aa4d-36f5694096eb", "evidence": "image/screenshot: Two side-by-side simulation windows with 'Left is better' and 'Right is better' buttons above them.", "pageRefs": null, "priority": null, "whenToUse": null, "confidence": 0.7, "extraction": { "at": "2026-07-16 23:23:09.050328+00", "model": "or:meta-llama/llama-4-scout", "runId": null, "seconds": null, "promptVersion": null }, "pageNumber": 10, "whyItWorks": null, "antipattern": null, "cardinality": "common", "narrativePurpose": null } ], "frameworks": [ { "matchId": "19c5ff67-57ca-4a5d-ade2-2718efa28e53", "evidence": "The slide depicts the binary choice mechanism central to RLHF.", "framework": null, "confidence": 0.95, "extraction": { "at": "2026-05-02 14:21:43.585+00", "model": "google/gemini-3.1-flash-lite-preview-20260303", "runId": null, "seconds": null, "promptVersion": null }, "pageNumber": 10, "frameworkName": "reinforcement-learning-from-human-feedback" } ], "arcBeats": [ { "to": 11, "arc": { "name": "Problem-Agitate-Solution", "slug": "problem-agitate-solution", "status": "active", "bestFor": "Sales pitches, urgency-driven, competitive displacement", "canonId": "019dd956-5ec6-70f9-9fd8-fdbaaff39acf", "version": 1, "bodyDocId": "019df22a-2356-73eb-8061-d8c81dc0aa49", "structure": "Problem (Identify pain) -> Agitate (Make it worse) -> Solution (Provide relief)", "description": null, "familyLabel": null }, "from": 8, "name": "Solution", "beatId": "72a1fd78-bc62-4d21-a603-262dd269d030", "matchId": "04c30632-8456-4dc0-8a7a-f7d40f1c761f", "beatType": { "name": "Solution (Provide relief)", "slug": "problem-agitate-solution-solution-provide-relief", "status": "active", "canonId": "019dd9b8-0fee-73e9-9ae5-799c7f3895cd", "version": 1, "description": null }, "evidence": "The deck proposes a solution using human feedback training processes and provides evidence to support it.", "position": 2, "isPrimaryArc": true, "parentBeatType": { "name": "Resolution", "slug": "resolution", "status": "active", "canonId": "019dd9b8-045d-703b-8dd7-e92aba3ba91b", "version": 1, "description": "How the deck ends. Required in 19/20 arcs. The exception (Sparkline) ends with new bliss which is functionally a resolution." }, "alignedBlockIds": null, "matchConfidence": 0.9 } ], "loops": [ { "to": 14, "from": 8, "loop": { "name": "53_jobs_to_be_done", "slug": "53-jobs-to-be-done", "status": "active", "bestFor": "Product innovation, market entry, competitive positioning", "canonId": "019dd956-7b6c-7594-b5b3-eecbe0ab48e8", "version": 1, "bodyDocId": "019df22a-2420-77be-bc41-ded96d08cb21", "structure": "The Customer's Job -> Current Solutions (Hired/Fired) -> Unmet Needs -> Our Solution Fit", "description": "Reframe the problem around what the customer is trying to accomplish, not what they're buying", "familyLabel": null }, "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?", "confidence": 0.7, "extraction": { "at": "2026-07-16 22:06:19.201225+00", "model": "or:meta-llama/llama-4-scout", "runId": null, "seconds": null, "promptVersion": "doc-narrative-v1" } } ], "locked": true }, { "page": 11, "type": "setup", "title": "The diagram shows a standard RL loop (RL Algorithm <-> Environment) augmented with a Reward Predictor that receives human feedback.", "function": "present_framework", "rawType": "propose_solution", "block": null, "metadata": { "slideType": "propose_solution", "slideTypeCanon": { "name": "Propose solution (contrarian)", "slug": "propose_solution", "status": "active", "canonId": "019de52d-1274-703c-81c6-5550f451697b", "version": 1, "description": null }, "function": "present_framework", "functionCanon": { "name": "Present framework", "slug": "present_framework", "status": "active", "canonId": "019de52d-1403-7667-9100-d2ac6709f750", "version": 1, "description": null }, "density": "balanced", "densityScore": 28, "componentCount": 3, "textChars": 274, "nDataPoints": 1, "notes": "The diagram shows a standard RL loop (RL Algorithm <-> Environment) augmented with a Reward Predictor that receives human feedback.", "elementsJson": null, "confidence": 1, "extraction": { "model": "google/gemini-3.1-flash-lite-preview-20260303", "runId": null, "at": "2026-05-02 14:21:41.665+00", "seconds": 2.717, "promptVersion": "v2-canon-conditioned" } }, "links": { "slide": "/slides/019de06f-84e4-70e4-bce4-3c7c4f09a8ea/11", "deck": "/decks/019de06f-84e4-70e4-bce4-3c7c4f09a8ea", "deckAnchor": "/decks/019de06f-84e4-70e4-bce4-3c7c4f09a8ea#slide-11", "deckJson": "/decks/019de06f-84e4-70e4-bce4-3c7c4f09a8ea.json" }, "components": [ { "bbox": { "h": 0.38, "w": 0.54, "x": 0.23, "y": 0.18 }, "kind": "diagram", "text": "RL loop with reward predictor", "attrs": null, "canon": { "kind": { "name": "Diagram", "slug": "diagram", "status": "active", "canonId": "019de52c-f9f0-734a-81de-15382ff28c65", "version": 1, "description": "Schematic or relational diagram." }, "tool": null, "subkind": { "name": "Process flow", "slug": "process", "status": "active", "canonId": "019de52d-039c-718f-ad3a-efe9c4b3c838", "version": 1, "description": null }, "framework": null }, "subkind": "process", "groupRole": null, "confidence": null, "extraction": { "at": "2026-05-02 14:21:41.665+00", "model": "google/gemini-3.1-flash-lite-preview-20260303", "runId": null, "seconds": null, "promptVersion": null }, "componentId": "7437ecdd-5d76-404f-a2c8-e52685ba95f2", "parentComponentId": null }, { "bbox": { "h": 0.28, "w": 0.88, "x": 0.06, "y": 0.64 }, "kind": "list", "text": "Bottom 2 boxes are standard reinforcement learning\nInjecting reward predictor modifies RL to learn the human's preference from sparse supervision (\"surrogate human\")\nRequest feedback on only 0.1% of experience; use active learning", "attrs": null, "canon": { "kind": { "name": "List", "slug": "list", "status": "active", "canonId": "019de52c-f94b-779b-a53b-fd4904a5f4d2", "version": 1, "description": "Bullet or numbered list." }, "tool": null, "subkind": { "name": "Bullet list", "slug": "bullet", "status": "active", "canonId": "019de52c-fc98-7169-a083-5cab805076ca", "version": 1, "description": null }, "framework": null }, "subkind": "bullet", "groupRole": null, "confidence": null, "extraction": { "at": "2026-05-02 14:21:41.665+00", "model": "google/gemini-3.1-flash-lite-preview-20260303", "runId": null, "seconds": null, "promptVersion": null }, "componentId": "191a97bf-b575-4b04-8030-9f11fbcf66cf", "parentComponentId": null }, { "bbox": { "h": 0.06, "w": 0.25, "x": 0.04, "y": 0.06 }, "kind": "title", "text": "Behind the hood", "attrs": null, "canon": { "kind": { "name": "Title", "slug": "title", "status": "active", "canonId": "019de52c-f7dd-76d9-b848-b1f12bd85cf4", "version": 1, "description": "Slide title or headline." }, "tool": null, "subkind": { "name": "Headline", "slug": "headline", "status": "active", "canonId": "019de52c-fae1-7463-9a7c-d60668eced0d", "version": 1, "description": null }, "framework": null }, "subkind": "headline", "groupRole": null, "confidence": null, "extraction": { "at": "2026-05-02 14:21:41.665+00", "model": "google/gemini-3.1-flash-lite-preview-20260303", "runId": null, "seconds": null, "promptVersion": null }, "componentId": "3d30c46e-594e-413a-a427-3e823c982edc", "parentComponentId": null } ], "metrics": [], "tools": [], "frameworks": [ { "matchId": "11d8debb-9c89-4b9c-984f-2bdce2f78139", "evidence": "Diagram shows RL algorithm, environment, and reward predictor with human feedback.", "framework": null, "confidence": 1, "extraction": { "at": "2026-05-02 14:21:41.665+00", "model": "google/gemini-3.1-flash-lite-preview-20260303", "runId": null, "seconds": null, "promptVersion": null }, "pageNumber": 11, "frameworkName": "Reinforcement Learning from Human Feedback (RLHF)" } ], "arcBeats": [ { "to": 11, "arc": { "name": "Problem-Agitate-Solution", "slug": "problem-agitate-solution", "status": "active", "bestFor": "Sales pitches, urgency-driven, competitive displacement", "canonId": "019dd956-5ec6-70f9-9fd8-fdbaaff39acf", "version": 1, "bodyDocId": "019df22a-2356-73eb-8061-d8c81dc0aa49", "structure": "Problem (Identify pain) -> Agitate (Make it worse) -> Solution (Provide relief)", "description": null, "familyLabel": null }, "from": 8, "name": "Solution", "beatId": "72a1fd78-bc62-4d21-a603-262dd269d030", "matchId": "04c30632-8456-4dc0-8a7a-f7d40f1c761f", "beatType": { "name": "Solution (Provide relief)", "slug": "problem-agitate-solution-solution-provide-relief", "status": "active", "canonId": "019dd9b8-0fee-73e9-9ae5-799c7f3895cd", "version": 1, "description": null }, "evidence": "The deck proposes a solution using human feedback training processes and provides evidence to support it.", "position": 2, "isPrimaryArc": true, "parentBeatType": { "name": "Resolution", "slug": "resolution", "status": "active", "canonId": "019dd9b8-045d-703b-8dd7-e92aba3ba91b", "version": 1, "description": "How the deck ends. Required in 19/20 arcs. The exception (Sparkline) ends with new bliss which is functionally a resolution." }, "alignedBlockIds": null, "matchConfidence": 0.9 } ], "loops": [ { "to": 14, "from": 8, "loop": { "name": "53_jobs_to_be_done", "slug": "53-jobs-to-be-done", "status": "active", "bestFor": "Product innovation, market entry, competitive positioning", "canonId": "019dd956-7b6c-7594-b5b3-eecbe0ab48e8", "version": 1, "bodyDocId": "019df22a-2420-77be-bc41-ded96d08cb21", "structure": "The Customer's Job -> Current Solutions (Hired/Fired) -> Unmet Needs -> Our Solution Fit", "description": "Reframe the problem around what the customer is trying to accomplish, not what they're buying", "familyLabel": null }, "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?", "confidence": 0.7, "extraction": { "at": "2026-07-16 22:06:19.201225+00", "model": "or:meta-llama/llama-4-scout", "runId": null, "seconds": null, "promptVersion": "doc-narrative-v1" } } ], "locked": true }, { "page": 12, "type": "evidence", "title": "The slide shows two screenshots of the same game (likely Enduro) with different agent strategies.", "function": "illustrate_case", "rawType": "before_after", "block": null, "metadata": { "slideType": "before_after", "slideTypeCanon": { "name": "Before / after", "slug": "before_after", "status": "active", "canonId": "019de52d-11fd-771e-bb45-e27f849bc560", "version": 1, "description": null }, "function": "illustrate_case", "functionCanon": { "name": "Illustrate case", "slug": "illustrate_case", "status": "active", "canonId": "019de52d-13b4-772c-bb73-1873612a7463", "version": 1, "description": null }, "density": "sparse", "densityScore": 13, "componentCount": 3, "textChars": 89, "nDataPoints": 0, "notes": "The slide shows two screenshots of the same game (likely Enduro) with different agent strategies.", "elementsJson": null, "confidence": 0.9, "extraction": { "model": "google/gemini-3.1-flash-lite-preview-20260303", "runId": null, "at": "2026-05-02 14:21:45.478+00", "seconds": 6.274, "promptVersion": "v2-canon-conditioned" } }, "links": { "slide": "/slides/019de06f-84e4-70e4-bce4-3c7c4f09a8ea/12", "deck": "/decks/019de06f-84e4-70e4-bce4-3c7c4f09a8ea", "deckAnchor": "/decks/019de06f-84e4-70e4-bce4-3c7c4f09a8ea#slide-12", "deckJson": "/decks/019de06f-84e4-70e4-bce4-3c7c4f09a8ea.json" }, "components": [ { "bbox": { "h": 0.6, "w": 0.35, "x": 0.1, "y": 0.25 }, "kind": "image", "text": "Gameplay screenshot 1", "attrs": null, "canon": { "kind": { "name": "Image", "slug": "image", "status": "active", "canonId": "019de52c-f978-701f-a720-51d029d769e7", "version": 1, "description": "Photo, illustration, screenshot or icon." }, "tool": null, "subkind": { "name": "Screenshot", "slug": "screenshot", "status": "active", "canonId": "019de52c-fd87-7509-8391-3d79fb339379", "version": 1, "description": null }, "framework": null }, "subkind": "screenshot", "groupRole": null, "confidence": null, "extraction": { "at": "2026-05-02 14:21:45.478+00", "model": "google/gemini-3.1-flash-lite-preview-20260303", "runId": null, "seconds": null, "promptVersion": null }, "componentId": "669a475f-0f39-4ccc-a460-464de7fec094", "parentComponentId": null }, { "bbox": { "h": 0.6, "w": 0.35, "x": 0.55, "y": 0.25 }, "kind": "image", "text": "Gameplay screenshot 2", "attrs": null, "canon": { "kind": { "name": "Image", "slug": "image", "status": "active", "canonId": "019de52c-f978-701f-a720-51d029d769e7", "version": 1, "description": "Photo, illustration, screenshot or icon." }, "tool": null, "subkind": { "name": "Screenshot", "slug": "screenshot", "status": "active", "canonId": "019de52c-fd87-7509-8391-3d79fb339379", "version": 1, "description": null }, "framework": null }, "subkind": "screenshot", "groupRole": null, "confidence": null, "extraction": { "at": "2026-05-02 14:21:45.478+00", "model": "google/gemini-3.1-flash-lite-preview-20260303", "runId": null, "seconds": null, "promptVersion": null }, "componentId": "8c0a9d27-dd6c-4ec7-beba-191bd4e560b3", "parentComponentId": null }, { "bbox": { "h": 0.1, "w": 0.9, "x": 0.05, "y": 0.05 }, "kind": "title", "text": "Training 2 different behaviors on the same game", "attrs": null, "canon": { "kind": { "name": "Title", "slug": "title", "status": "active", "canonId": "019de52c-f7dd-76d9-b848-b1f12bd85cf4", "version": 1, "description": "Slide title or headline." }, "tool": null, "subkind": { "name": "Headline", "slug": "headline", "status": "active", "canonId": "019de52c-fae1-7463-9a7c-d60668eced0d", "version": 1, "description": null }, "framework": null }, "subkind": "headline", "groupRole": null, "confidence": null, "extraction": { "at": "2026-05-02 14:21:45.478+00", "model": "google/gemini-3.1-flash-lite-preview-20260303", "runId": null, "seconds": null, "promptVersion": null }, "componentId": "75303c0b-668d-4817-9065-246f454c5c06", "parentComponentId": null } ], "metrics": [], "tools": [ { "tool": { "name": "Before-After-Bridge", "slug": "before-after-bridge", "status": "active", "bestFor": null, "canonId": "019dd956-8604-7188-883f-42370288ad05", "version": 1, "bodyDocId": null, "description": "Narrative structure: Paint the before, show the after, explain the bridge", "familyLabel": null, "categoryName": "Block", "categorySlug": "block" }, "agent": "Storyteller", "layer": "slide", "agents": [ "Storyteller" ], "matchId": "47e48a5c-c1cf-4d0a-8da6-32c8a3163bdd", "evidence": "The slide shows two gameplay screenshots with a headline 'Training 2 different behaviors on the same game'.", "pageRefs": null, "priority": null, "whenToUse": null, "confidence": 0.8, "extraction": { "at": "2026-07-16 23:22:57.988001+00", "model": "or:meta-llama/llama-4-scout", "runId": null, "seconds": null, "promptVersion": null }, "pageNumber": 12, "whyItWorks": null, "antipattern": null, "cardinality": null, "narrativePurpose": null } ], "frameworks": [ { "matchId": "45f2b249-2575-42fd-9bcc-995502dd0c59", "evidence": "The slide presents two distinct outcomes (behaviors) from the same input (game) to illustrate a point.", "framework": null, "confidence": 0.8, "extraction": { "at": "2026-05-02 14:21:45.478+00", "model": "google/gemini-3.1-flash-lite-preview-20260303", "runId": null, "seconds": null, "promptVersion": null }, "pageNumber": 12, "frameworkName": "before-after-framing" } ], "arcBeats": [], "loops": [ { "to": 14, "from": 8, "loop": { "name": "53_jobs_to_be_done", "slug": "53-jobs-to-be-done", "status": "active", "bestFor": "Product innovation, market entry, competitive positioning", "canonId": "019dd956-7b6c-7594-b5b3-eecbe0ab48e8", "version": 1, "bodyDocId": "019df22a-2420-77be-bc41-ded96d08cb21", "structure": "The Customer's Job -> Current Solutions (Hired/Fired) -> Unmet Needs -> Our Solution Fit", "description": "Reframe the problem around what the customer is trying to accomplish, not what they're buying", "familyLabel": null }, "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?", "confidence": 0.7, "extraction": { "at": "2026-07-16 22:06:19.201225+00", "model": "or:meta-llama/llama-4-scout", "runId": null, "seconds": null, "promptVersion": "doc-narrative-v1" } } ], "locked": true }, { "page": 13, "type": "appendix", "title": "The images show Breakout, Pong, and Seaquest.", "function": "other", "rawType": "other", "block": null, "metadata": { "slideType": "other", "slideTypeCanon": { "name": "Other (fallback)", "slug": "other", "status": "active", "canonId": "019de52d-0af5-71ed-83de-d51ae245fc43", "version": 1, "description": null }, "function": "other", "functionCanon": { "name": "Other (fallback)", "slug": "other", "status": "active", "canonId": "019de52d-15e1-75bc-99dc-07732380e2ce", "version": 1, "description": null }, "density": "sparse", "densityScore": 22, "componentCount": 4, "textChars": 78, "nDataPoints": 0, "notes": "The images show Breakout, Pong, and Seaquest.", "elementsJson": null, "confidence": 0.9, "extraction": { "model": "google/gemini-3.1-flash-lite-preview-20260303", "runId": null, "at": "2026-05-02 14:21:42.151+00", "seconds": 2.513, "promptVersion": "v2-canon-conditioned" } }, "links": { "slide": "/slides/019de06f-84e4-70e4-bce4-3c7c4f09a8ea/13", "deck": "/decks/019de06f-84e4-70e4-bce4-3c7c4f09a8ea", "deckAnchor": "/decks/019de06f-84e4-70e4-bce4-3c7c4f09a8ea#slide-13", "deckJson": "/decks/019de06f-84e4-70e4-bce4-3c7c4f09a8ea.json" }, "components": [ { "bbox": { "h": 0.65, "w": 0.3, "x": 0.35, "y": 0.28 }, "kind": "image", "text": "Pong game screenshot", "attrs": null, "canon": { "kind": { "name": "Image", "slug": "image", "status": "active", "canonId": "019de52c-f978-701f-a720-51d029d769e7", "version": 1, "description": "Photo, illustration, screenshot or icon." }, "tool": null, "subkind": { "name": "Screenshot", "slug": "screenshot", "status": "active", "canonId": "019de52c-fd87-7509-8391-3d79fb339379", "version": 1, "description": null }, "framework": null }, "subkind": "screenshot", "groupRole": null, "confidence": null, "extraction": { "at": "2026-05-02 14:21:42.151+00", "model": "google/gemini-3.1-flash-lite-preview-20260303", "runId": null, "seconds": null, "promptVersion": null }, "componentId": "523a4c4a-4dbe-48ef-ae5c-b195766a32f0", "parentComponentId": null }, { "bbox": { "h": 0.65, "w": 0.3, "x": 0.02, "y": 0.28 }, "kind": "image", "text": "Breakout game screenshot", "attrs": null, "canon": { "kind": { "name": "Image", "slug": "image", "status": "active", "canonId": "019de52c-f978-701f-a720-51d029d769e7", "version": 1, "description": "Photo, illustration, screenshot or icon." }, "tool": null, "subkind": { "name": "Screenshot", "slug": "screenshot", "status": "active", "canonId": "019de52c-fd87-7509-8391-3d79fb339379", "version": 1, "description": null }, "framework": null }, "subkind": "screenshot", "groupRole": null, "confidence": null, "extraction": { "at": "2026-05-02 14:21:42.151+00", "model": "google/gemini-3.1-flash-lite-preview-20260303", "runId": null, "seconds": null, "promptVersion": null }, "componentId": "6a669c43-1021-47ec-a6ad-2a6e453ad342", "parentComponentId": null }, { "bbox": { "h": 0.65, "w": 0.3, "x": 0.68, "y": 0.28 }, "kind": "image", "text": "Seaquest game screenshot", "attrs": null, "canon": { "kind": { "name": "Image", "slug": "image", "status": "active", "canonId": "019de52c-f978-701f-a720-51d029d769e7", "version": 1, "description": "Photo, illustration, screenshot or icon." }, "tool": null, "subkind": { "name": "Screenshot", "slug": "screenshot", "status": "active", "canonId": "019de52c-fd87-7509-8391-3d79fb339379", "version": 1, "description": null }, "framework": null }, "subkind": "screenshot", "groupRole": null, "confidence": null, "extraction": { "at": "2026-05-02 14:21:42.151+00", "model": "google/gemini-3.1-flash-lite-preview-20260303", "runId": null, "seconds": null, "promptVersion": null }, "componentId": "71d8a786-b3c8-4617-980c-c9892a89720d", "parentComponentId": null }, { "bbox": { "h": 0.08, "w": 0.2, "x": 0.04, "y": 0.08 }, "kind": "title", "text": "More Tasks", "attrs": null, "canon": { "kind": { "name": "Title", "slug": "title", "status": "active", "canonId": "019de52c-f7dd-76d9-b848-b1f12bd85cf4", "version": 1, "description": "Slide title or headline." }, "tool": null, "subkind": { "name": "Headline", "slug": "headline", "status": "active", "canonId": "019de52c-fae1-7463-9a7c-d60668eced0d", "version": 1, "description": null }, "framework": null }, "subkind": "headline", "groupRole": null, "confidence": null, "extraction": { "at": "2026-05-02 14:21:42.151+00", "model": "google/gemini-3.1-flash-lite-preview-20260303", "runId": null, "seconds": null, "promptVersion": null }, "componentId": "831c58ec-93a8-42cd-871a-00aab4e42885", "parentComponentId": null } ], "metrics": [], "tools": [], "frameworks": [], "arcBeats": [], "loops": [ { "to": 14, "from": 8, "loop": { "name": "53_jobs_to_be_done", "slug": "53-jobs-to-be-done", "status": "active", "bestFor": "Product innovation, market entry, competitive positioning", "canonId": "019dd956-7b6c-7594-b5b3-eecbe0ab48e8", "version": 1, "bodyDocId": "019df22a-2420-77be-bc41-ded96d08cb21", "structure": "The Customer's Job -> Current Solutions (Hired/Fired) -> Unmet Needs -> Our Solution Fit", "description": "Reframe the problem around what the customer is trying to accomplish, not what they're buying", "familyLabel": null }, "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?", "confidence": 0.7, "extraction": { "at": "2026-07-16 22:06:19.201225+00", "model": "or:meta-llama/llama-4-scout", "runId": null, "seconds": null, "promptVersion": "doc-narrative-v1" } } ], "locked": true }, { "page": 14, "type": "analysis", "title": "The charts show reward over time (timesteps) for different training methodologies.", "function": "analyze_data", "rawType": "kpi_overview", "block": null, "metadata": { "slideType": "kpi_overview", "slideTypeCanon": { "name": "KPI overview (contrarian)", "slug": "kpi_overview", "status": "active", "canonId": "019de52d-11d5-7328-b1d1-5368a03e7796", "version": 1, "description": null }, "function": "analyze_data", "functionCanon": { "name": "Analyze data", "slug": "analyze_data", "status": "active", "canonId": "019de52d-1590-7413-8c82-cc218cff40b4", "version": 1, "description": null }, "density": "balanced", "densityScore": 26, "componentCount": 3, "textChars": 198, "nDataPoints": 7, "notes": "The charts show reward over time (timesteps) for different training methodologies.", "elementsJson": null, "confidence": 1, "extraction": { "model": "google/gemini-3.1-flash-lite-preview-20260303", "runId": null, "at": "2026-05-02 14:21:46.713+00", "seconds": 7.035, "promptVersion": "v2-canon-conditioned" } }, "links": { "slide": "/slides/019de06f-84e4-70e4-bce4-3c7c4f09a8ea/14", "deck": "/decks/019de06f-84e4-70e4-bce4-3c7c4f09a8ea", "deckAnchor": "/decks/019de06f-84e4-70e4-bce4-3c7c4f09a8ea#slide-14", "deckJson": "/decks/019de06f-84e4-70e4-bce4-3c7c4f09a8ea.json" }, "components": [ { "bbox": { "h": 0.75, "w": 0.85, "x": 0.08, "y": 0.18 }, "kind": "chart", "text": "Performance charts for beamrider, breakout, pong, qbert, seaquest, spaceinv, enduro", "attrs": null, "canon": { "kind": { "name": "Chart", "slug": "chart", "status": "active", "canonId": "019de52c-f9a0-76ef-adc6-0b859109d1f8", "version": 1, "description": "Quantitative chart." }, "tool": null, "subkind": { "name": "Line chart", "slug": "line", "status": "active", "canonId": "019de52c-ff62-778f-8f6c-a952ceae8677", "version": 1, "description": null }, "framework": null }, "subkind": "line", "groupRole": null, "confidence": null, "extraction": { "at": "2026-05-02 14:21:46.713+00", "model": "google/gemini-3.1-flash-lite-preview-20260303", "runId": null, "seconds": null, "promptVersion": null }, "componentId": "cdaf92d9-d82a-4500-966e-0a254c5be7e4", "parentComponentId": null }, { "bbox": { "h": 0.15, "w": 0.15, "x": 0.75, "y": 0.7 }, "kind": "legend", "text": "RL, 10k synthetic labels, 5.6k synthetic labels, 3.3k synthetic labels, 5.5k human labels", "attrs": null, "canon": { "kind": { "name": "Legend", "slug": "legend", "status": "active", "canonId": "31361933-3d22-4c70-bbad-0801e9fffc0c", "version": 1, "description": "Color/symbol decoder placed adjacent to a chart or diagram." }, "tool": null, "subkind": null, "framework": null }, "subkind": null, "groupRole": null, "confidence": null, "extraction": { "at": "2026-05-02 14:21:46.713+00", "model": "google/gemini-3.1-flash-lite-preview-20260303", "runId": null, "seconds": null, "promptVersion": null }, "componentId": "262f03e7-4c17-4a56-b0f2-b26400bdca80", "parentComponentId": null }, { "bbox": { "h": 0.08, "w": 0.4, "x": 0.05, "y": 0.05 }, "kind": "title", "text": "Performance on Atari Games", "attrs": null, "canon": { "kind": { "name": "Title", "slug": "title", "status": "active", "canonId": "019de52c-f7dd-76d9-b848-b1f12bd85cf4", "version": 1, "description": "Slide title or headline." }, "tool": null, "subkind": { "name": "Headline", "slug": "headline", "status": "active", "canonId": "019de52c-fae1-7463-9a7c-d60668eced0d", "version": 1, "description": null }, "framework": null }, "subkind": "headline", "groupRole": null, "confidence": null, "extraction": { "at": "2026-05-02 14:21:46.713+00", "model": "google/gemini-3.1-flash-lite-preview-20260303", "runId": null, "seconds": null, "promptVersion": null }, "componentId": "a8f1a791-e78d-4fbf-a619-99c74f9aa746", "parentComponentId": null } ], "metrics": [], "tools": [], "frameworks": [], "arcBeats": [], "loops": [ { "to": 14, "from": 8, "loop": { "name": "53_jobs_to_be_done", "slug": "53-jobs-to-be-done", "status": "active", "bestFor": "Product innovation, market entry, competitive positioning", "canonId": "019dd956-7b6c-7594-b5b3-eecbe0ab48e8", "version": 1, "bodyDocId": "019df22a-2420-77be-bc41-ded96d08cb21", "structure": "The Customer's Job -> Current Solutions (Hired/Fired) -> Unmet Needs -> Our Solution Fit", "description": "Reframe the problem around what the customer is trying to accomplish, not what they're buying", "familyLabel": null }, "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?", "confidence": 0.7, "extraction": { "at": "2026-07-16 22:06:19.201225+00", "model": "or:meta-llama/llama-4-scout", "runId": null, "seconds": null, "promptVersion": "doc-narrative-v1" } } ], "locked": true }, { "page": 15, "type": "analysis", "title": "The charts show learning curves for different training methods. The legend is consistent across all sub-plots.", "function": "analyze_data", "rawType": "kpi_overview", "block": null, "metadata": { "slideType": "kpi_overview", "slideTypeCanon": { "name": "KPI overview (contrarian)", "slug": "kpi_overview", "status": "active", "canonId": "019de52d-11d5-7328-b1d1-5368a03e7796", "version": 1, "description": null }, "function": "analyze_data", "functionCanon": { "name": "Analyze data", "slug": "analyze_data", "status": "active", "canonId": "019de52d-1590-7413-8c82-cc218cff40b4", "version": 1, "description": null }, "density": "balanced", "densityScore": 26, "componentCount": 3, "textChars": 151, "nDataPoints": 35, "notes": "The charts show learning curves for different training methods. The legend is consistent across all sub-plots.", "elementsJson": null, "confidence": 1, "extraction": { "model": "google/gemini-3.1-flash-lite-preview-20260303", "runId": null, "at": "2026-05-02 14:21:45.163+00", "seconds": 5.347, "promptVersion": "v2-canon-conditioned" } }, "links": { "slide": "/slides/019de06f-84e4-70e4-bce4-3c7c4f09a8ea/15", "deck": "/decks/019de06f-84e4-70e4-bce4-3c7c4f09a8ea", "deckAnchor": "/decks/019de06f-84e4-70e4-bce4-3c7c4f09a8ea#slide-15", "deckJson": "/decks/019de06f-84e4-70e4-bce4-3c7c4f09a8ea.json" }, "components": [ { "bbox": { "h": 0.75, "w": 0.85, "x": 0.08, "y": 0.18 }, "kind": "chart", "text": "Performance charts for 7 Atari games", "attrs": null, "canon": { "kind": { "name": "Chart", "slug": "chart", "status": "active", "canonId": "019de52c-f9a0-76ef-adc6-0b859109d1f8", "version": 1, "description": "Quantitative chart." }, "tool": null, "subkind": { "name": "Line chart", "slug": "line", "status": "active", "canonId": "019de52c-ff62-778f-8f6c-a952ceae8677", "version": 1, "description": null }, "framework": null }, "subkind": "line", "groupRole": null, "confidence": null, "extraction": { "at": "2026-05-02 14:21:45.163+00", "model": "google/gemini-3.1-flash-lite-preview-20260303", "runId": null, "seconds": null, "promptVersion": null }, "componentId": "72043249-b1bf-43d4-aece-e790ca7a84b8", "parentComponentId": null }, { "bbox": { "h": 0.15, "w": 0.15, "x": 0.78, "y": 0.75 }, "kind": "legend", "text": "RL, 10k synthetic labels, 5.6k synthetic labels, 3.3k synthetic labels, 5.5k human labels", "attrs": null, "canon": { "kind": { "name": "Legend", "slug": "legend", "status": "active", "canonId": "31361933-3d22-4c70-bbad-0801e9fffc0c", "version": 1, "description": "Color/symbol decoder placed adjacent to a chart or diagram." }, "tool": null, "subkind": null, "framework": null }, "subkind": null, "groupRole": null, "confidence": null, "extraction": { "at": "2026-05-02 14:21:45.163+00", "model": "google/gemini-3.1-flash-lite-preview-20260303", "runId": null, "seconds": null, "promptVersion": null }, "componentId": "493b4278-676f-4f30-b14a-94e6f762dde3", "parentComponentId": null }, { "bbox": { "h": 0.08, "w": 0.4, "x": 0.05, "y": 0.05 }, "kind": "title", "text": "Performance on Atari Games", "attrs": null, "canon": { "kind": { "name": "Title", "slug": "title", "status": "active", "canonId": "019de52c-f7dd-76d9-b848-b1f12bd85cf4", "version": 1, "description": "Slide title or headline." }, "tool": null, "subkind": { "name": "Headline", "slug": "headline", "status": "active", "canonId": "019de52c-fae1-7463-9a7c-d60668eced0d", "version": 1, "description": null }, "framework": null }, "subkind": "headline", "groupRole": null, "confidence": null, "extraction": { "at": "2026-05-02 14:21:45.163+00", "model": "google/gemini-3.1-flash-lite-preview-20260303", "runId": null, "seconds": null, "promptVersion": null }, "componentId": "194e6120-ecfa-467d-933e-efbd118a262e", "parentComponentId": null } ], "metrics": [], "tools": [], "frameworks": [], "arcBeats": [], "loops": [], "locked": true }, { "page": 16, "type": "analysis", "title": "The charts show learning curves for reinforcement learning agents. The legend is consistent across all plots.", "function": "analyze_data", "rawType": "appendix_data", "block": null, "metadata": { "slideType": "appendix_data", "slideTypeCanon": { "name": "Appendix data", "slug": "appendix_data", "status": "active", "canonId": "019de52d-078a-745b-9ee4-c15f2f539acd", "version": 1, "description": null }, "function": "analyze_data", "functionCanon": { "name": "Analyze data", "slug": "analyze_data", "status": "active", "canonId": "019de52d-1590-7413-8c82-cc218cff40b4", "version": 1, "description": null }, "density": "balanced", "densityScore": 30, "componentCount": 3, "textChars": 263, "nDataPoints": 8, "notes": "The charts show learning curves for reinforcement learning agents. The legend is consistent across all plots.", "elementsJson": null, "confidence": 1, "extraction": { "model": "google/gemini-3.1-flash-lite-preview-20260303", "runId": null, "at": "2026-05-02 14:21:46.441+00", "seconds": 6.598, "promptVersion": "v2-canon-conditioned" } }, "links": { "slide": "/slides/019de06f-84e4-70e4-bce4-3c7c4f09a8ea/16", "deck": "/decks/019de06f-84e4-70e4-bce4-3c7c4f09a8ea", "deckAnchor": "/decks/019de06f-84e4-70e4-bce4-3c7c4f09a8ea#slide-16", "deckJson": "/decks/019de06f-84e4-70e4-bce4-3c7c4f09a8ea.json" }, "components": [ { "bbox": { "h": 0.8, "w": 0.8, "x": 0.1, "y": 0.15 }, "kind": "chart", "text": "Grid of 8 line charts showing reward vs timestep for walker, hopper, swimmer, cheetah, ant, reacher, double-pendulum, and pendulum.", "attrs": null, "canon": { "kind": { "name": "Chart", "slug": "chart", "status": "active", "canonId": "019de52c-f9a0-76ef-adc6-0b859109d1f8", "version": 1, "description": "Quantitative chart." }, "tool": null, "subkind": { "name": "Line chart", "slug": "line", "status": "active", "canonId": "019de52c-ff62-778f-8f6c-a952ceae8677", "version": 1, "description": null }, "framework": null }, "subkind": "line", "groupRole": null, "confidence": null, "extraction": { "at": "2026-05-02 14:21:46.441+00", "model": "google/gemini-3.1-flash-lite-preview-20260303", "runId": null, "seconds": null, "promptVersion": null }, "componentId": "557c962a-ca99-4012-a850-1695fa4ba33c", "parentComponentId": null }, { "bbox": { "h": 0.15, "w": 0.15, "x": 0.75, "y": 0.75 }, "kind": "legend", "text": "Legend: real reward, 1400 synthetic queries, 700 synthetic queries, 350 synthetic queries, 750 human queries", "attrs": null, "canon": { "kind": { "name": "Legend", "slug": "legend", "status": "active", "canonId": "31361933-3d22-4c70-bbad-0801e9fffc0c", "version": 1, "description": "Color/symbol decoder placed adjacent to a chart or diagram." }, "tool": null, "subkind": null, "framework": null }, "subkind": null, "groupRole": null, "confidence": null, "extraction": { "at": "2026-05-02 14:21:46.441+00", "model": "google/gemini-3.1-flash-lite-preview-20260303", "runId": null, "seconds": null, "promptVersion": null }, "componentId": "d1338fc3-2269-46ab-a7ea-72a91e299388", "parentComponentId": null }, { "bbox": { "h": 0.05, "w": 0.3, "x": 0.05, "y": 0.05 }, "kind": "title", "text": "Simulated Robotics Tasks", "attrs": null, "canon": { "kind": { "name": "Title", "slug": "title", "status": "active", "canonId": "019de52c-f7dd-76d9-b848-b1f12bd85cf4", "version": 1, "description": "Slide title or headline." }, "tool": null, "subkind": { "name": "Headline", "slug": "headline", "status": "active", "canonId": "019de52c-fae1-7463-9a7c-d60668eced0d", "version": 1, "description": null }, "framework": null }, "subkind": "headline", "groupRole": null, "confidence": null, "extraction": { "at": "2026-05-02 14:21:46.441+00", "model": "google/gemini-3.1-flash-lite-preview-20260303", "runId": null, "seconds": null, "promptVersion": null }, "componentId": "7869bf03-befd-42a1-b40c-fd501cfbddac", "parentComponentId": null } ], "metrics": [], "tools": [], "frameworks": [], "arcBeats": [], "loops": [], "locked": true }, { "page": 17, "type": "appendix", "title": "The slide uses a reinforcement learning-style feedback loop diagram (Agent-Environment interaction) to explain how human intent is translated into technical act", "function": "present_framework", "rawType": "other", "block": null, "metadata": { "slideType": "other", "slideTypeCanon": { "name": "Other (fallback)", "slug": "other", "status": "active", "canonId": "019de52d-0af5-71ed-83de-d51ae245fc43", "version": 1, "description": null }, "function": "present_framework", "functionCanon": { "name": "Present framework", "slug": "present_framework", "status": "active", "canonId": "019de52d-1403-7667-9100-d2ac6709f750", "version": 1, "description": null }, "density": "dense", "densityScore": 56, "componentCount": 7, "textChars": 230, "nDataPoints": 0, "notes": "The slide uses a reinforcement learning-style feedback loop diagram (Agent-Environment interaction) to explain how human intent is translated into technical actions.", "elementsJson": null, "confidence": 0.9, "extraction": { "model": "google/gemini-3.1-flash-lite-preview-20260303", "runId": null, "at": "2026-05-02 14:21:42.642+00", "seconds": 2.772, "promptVersion": "v2-canon-conditioned" } }, "links": { "slide": "/slides/019de06f-84e4-70e4-bce4-3c7c4f09a8ea/17", "deck": "/decks/019de06f-84e4-70e4-bce4-3c7c4f09a8ea", "deckAnchor": "/decks/019de06f-84e4-70e4-bce4-3c7c4f09a8ea#slide-17", "deckJson": "/decks/019de06f-84e4-70e4-bce4-3c7c4f09a8ea.json" }, "components": [ { "bbox": { "h": 0.8, "w": 0.9, "x": 0.05, "y": 0.15 }, "kind": "diagram", "text": "Agent-Environment interaction loop", "attrs": null, "canon": { "kind": { "name": "Diagram", "slug": "diagram", "status": "active", "canonId": "019de52c-f9f0-734a-81de-15382ff28c65", "version": 1, "description": "Schematic or relational diagram." }, "tool": null, "subkind": { "name": "Process flow", "slug": "process", "status": "active", "canonId": "019de52d-039c-718f-ad3a-efe9c4b3c838", "version": 1, "description": null }, "framework": null }, "subkind": "process", "groupRole": null, "confidence": null, "extraction": { "at": "2026-05-02 14:21:42.642+00", "model": "google/gemini-3.1-flash-lite-preview-20260303", "runId": null, "seconds": null, "promptVersion": null }, "componentId": "90be7c96-346a-4586-825f-e8c432c1cd2a", "parentComponentId": null }, { "bbox": { "h": 0.15, "w": 0.2, "x": 0.09, "y": 0.65 }, "kind": "image", "text": "Brain icon", "attrs": null, "canon": { "kind": { "name": "Image", "slug": "image", "status": "active", "canonId": "019de52c-f978-701f-a720-51d029d769e7", "version": 1, "description": "Photo, illustration, screenshot or icon." }, "tool": null, "subkind": { "name": "Illustration", "slug": "illustration", "status": "active", "canonId": "019de52c-fe76-7772-a5ac-c0b7294d7615", "version": 1, "description": null }, "framework": null }, "subkind": "illustration", "groupRole": null, "confidence": null, "extraction": { "at": "2026-05-02 14:21:42.642+00", "model": "google/gemini-3.1-flash-lite-preview-20260303", "runId": null, "seconds": null, "promptVersion": null }, "componentId": "9c231adf-d889-4d68-b644-59624e72599c", "parentComponentId": null }, { "bbox": { "h": 0.2, "w": 0.2, "x": 0.75, "y": 0.55 }, "kind": "image", "text": "Earth image", "attrs": null, "canon": { "kind": { "name": "Image", "slug": "image", "status": "active", "canonId": "019de52c-f978-701f-a720-51d029d769e7", "version": 1, "description": "Photo, illustration, screenshot or icon." }, "tool": null, "subkind": { "name": "Illustration", "slug": "illustration", "status": "active", "canonId": "019de52c-fe76-7772-a5ac-c0b7294d7615", "version": 1, "description": null }, "framework": null }, "subkind": "illustration", "groupRole": null, "confidence": null, "extraction": { "at": "2026-05-02 14:21:42.642+00", "model": "google/gemini-3.1-flash-lite-preview-20260303", "runId": null, "seconds": null, "promptVersion": null }, "componentId": "acac6446-f352-49d7-b47a-996b950ad06d", "parentComponentId": null }, { "bbox": { "h": 0.2, "w": 0.15, "x": 0.55, "y": 0.05 }, "kind": "image", "text": "Human stick figure", "attrs": null, "canon": { "kind": { "name": "Image", "slug": "image", "status": "active", "canonId": "019de52c-f978-701f-a720-51d029d769e7", "version": 1, "description": "Photo, illustration, screenshot or icon." }, "tool": null, "subkind": { "name": "Illustration", "slug": "illustration", "status": "active", "canonId": "019de52c-fe76-7772-a5ac-c0b7294d7615", "version": 1, "description": null }, "framework": null }, "subkind": "illustration", "groupRole": null, "confidence": null, "extraction": { "at": "2026-05-02 14:21:42.642+00", "model": "google/gemini-3.1-flash-lite-preview-20260303", "runId": null, "seconds": null, "promptVersion": null }, "componentId": "af7d9ac5-1d2a-4c4a-b884-2459857fd6ea", "parentComponentId": null }, { "bbox": { "h": 0.15, "w": 0.2, "x": 0.05, "y": 0.25 }, "kind": "image", "text": "Language interpreter word cloud", "attrs": null, "canon": { "kind": { "name": "Image", "slug": "image", "status": "active", "canonId": "019de52c-f978-701f-a720-51d029d769e7", "version": 1, "description": "Photo, illustration, screenshot or icon." }, "tool": null, "subkind": { "name": "Illustration", "slug": "illustration", "status": "active", "canonId": "019de52c-fe76-7772-a5ac-c0b7294d7615", "version": 1, "description": null }, "framework": null }, "subkind": "illustration", "groupRole": null, "confidence": null, "extraction": { "at": "2026-05-02 14:21:42.642+00", "model": "google/gemini-3.1-flash-lite-preview-20260303", "runId": null, "seconds": null, "promptVersion": null }, "componentId": "e630dfdf-897d-4ad5-927f-b0353f7423b6", "parentComponentId": null }, { "bbox": { "h": 0.1, "w": 0.3, "x": 0.35, "y": 0.35 }, "kind": "paragraph", "text": "What is most important to you to prioritize? Extra safe or fast? How many people? Will they come back or stay?", "attrs": null, "canon": { "kind": { "name": "Paragraph", "slug": "paragraph", "status": "active", "canonId": "5f2affc0-941d-4d07-959d-4bfc1930cea4", "version": 1, "description": "Single prose block in the body of a slide." }, "tool": null, "subkind": { "name": "Paragraph", "slug": "paragraph", "status": "active", "canonId": "019de52c-fb80-75af-aa5c-a7a4bfac2433", "version": 1, "description": null }, "framework": null }, "subkind": "paragraph", "groupRole": null, "confidence": null, "extraction": { "at": "2026-05-02 14:21:42.642+00", "model": "google/gemini-3.1-flash-lite-preview-20260303", "runId": null, "seconds": null, "promptVersion": null }, "componentId": "5c67e6bd-b4e5-4cfe-9378-abe3fdb28de9", "parentComponentId": null }, { "bbox": { "h": 0.08, "w": 0.3, "x": 0.05, "y": 0.05 }, "kind": "title", "text": "Long Term Vision", "attrs": null, "canon": { "kind": { "name": "Title", "slug": "title", "status": "active", "canonId": "019de52c-f7dd-76d9-b848-b1f12bd85cf4", "version": 1, "description": "Slide title or headline." }, "tool": null, "subkind": { "name": "Headline", "slug": "headline", "status": "active", "canonId": "019de52c-fae1-7463-9a7c-d60668eced0d", "version": 1, "description": null }, "framework": null }, "subkind": "headline", "groupRole": null, "confidence": null, "extraction": { "at": "2026-05-02 14:21:42.642+00", "model": "google/gemini-3.1-flash-lite-preview-20260303", "runId": null, "seconds": null, "promptVersion": null }, "componentId": "40185e4f-4ec0-458e-86cb-298a4275c2ef", "parentComponentId": null } ], "metrics": [], "tools": [], "frameworks": [ { "matchId": "f9b7a512-2e79-4e13-849b-f1931b525a64", "evidence": "The diagram explicitly labels 'Agent' and 'Environment' with state transitions (ot, ot+1) and action (at).", "framework": null, "confidence": 0.95, "extraction": { "at": "2026-05-02 14:21:42.642+00", "model": "google/gemini-3.1-flash-lite-preview-20260303", "runId": null, "seconds": null, "promptVersion": null }, "pageNumber": 17, "frameworkName": "Reinforcement Learning Agent-Environment Loop" } ], "arcBeats": [], "loops": [], "locked": true }, { "page": 18, "type": "evidence", "title": "The slide uses a screenshot of a 2017 Economist article as a cultural/media reference to establish the context of the 'AI policy problem'.", "function": "cite_precedent", "rawType": "cite_precedent", "block": null, "metadata": { "slideType": "cite_precedent", "slideTypeCanon": { "name": "Cite precedent", "slug": "cite_precedent", "status": "active", "canonId": "019de52d-12ec-72be-8881-51b438fa76c7", "version": 1, "description": null }, "function": "cite_precedent", "functionCanon": { "name": "Cite precedent", "slug": "cite_precedent", "status": "active", "canonId": "019de52d-142b-714b-86d4-58aad7b568ec", "version": 1, "description": null }, "density": "sparse", "densityScore": 8, "componentCount": 2, "textChars": 97, "nDataPoints": 0, "notes": "The slide uses a screenshot of a 2017 Economist article as a cultural/media reference to establish the context of the 'AI policy problem'.", "elementsJson": null, "confidence": 0.9, "extraction": { "model": "google/gemini-3.1-flash-lite-preview-20260303", "runId": null, "at": "2026-05-02 14:21:44.729+00", "seconds": 4.844, "promptVersion": "v2-canon-conditioned" } }, "links": { "slide": "/slides/019de06f-84e4-70e4-bce4-3c7c4f09a8ea/18", "deck": "/decks/019de06f-84e4-70e4-bce4-3c7c4f09a8ea", "deckAnchor": "/decks/019de06f-84e4-70e4-bce4-3c7c4f09a8ea#slide-18", "deckJson": "/decks/019de06f-84e4-70e4-bce4-3c7c4f09a8ea.json" }, "components": [ { "bbox": { "h": 0.75, "w": 0.6, "x": 0.2, "y": 0.15 }, "kind": "image", "text": "Screenshot of The Economist article 'Why China's AI push is worrying'", "attrs": null, "canon": { "kind": { "name": "Image", "slug": "image", "status": "active", "canonId": "019de52c-f978-701f-a720-51d029d769e7", "version": 1, "description": "Photo, illustration, screenshot or icon." }, "tool": null, "subkind": { "name": "Screenshot", "slug": "screenshot", "status": "active", "canonId": "019de52c-fd87-7509-8391-3d79fb339379", "version": 1, "description": null }, "framework": null }, "subkind": "screenshot", "groupRole": null, "confidence": null, "extraction": { "at": "2026-05-02 14:21:44.729+00", "model": "google/gemini-3.1-flash-lite-preview-20260303", "runId": null, "seconds": null, "promptVersion": null }, "componentId": "40a9a92c-9d01-444f-b544-f62b3d5fff1c", "parentComponentId": null }, { "bbox": { "h": 0.05, "w": 0.4, "x": 0.05, "y": 0.05 }, "kind": "title", "text": "The Harder Problem of Policy", "attrs": null, "canon": { "kind": { "name": "Title", "slug": "title", "status": "active", "canonId": "019de52c-f7dd-76d9-b848-b1f12bd85cf4", "version": 1, "description": "Slide title or headline." }, "tool": null, "subkind": { "name": "Headline", "slug": "headline", "status": "active", "canonId": "019de52c-fae1-7463-9a7c-d60668eced0d", "version": 1, "description": null }, "framework": null }, "subkind": "headline", "groupRole": null, "confidence": null, "extraction": { "at": "2026-05-02 14:21:44.729+00", "model": "google/gemini-3.1-flash-lite-preview-20260303", "runId": null, "seconds": null, "promptVersion": null }, "componentId": "d6c85fe1-4cce-43ad-899f-859eaf855813", "parentComponentId": null } ], "metrics": [], "tools": [], "frameworks": [ { "matchId": "d06b1117-3761-4f07-b12a-2d142e6faced", "evidence": "Uses a reputable media source to validate the premise of the problem.", "framework": null, "confidence": 0.8, "extraction": { "at": "2026-05-02 14:21:44.729+00", "model": "google/gemini-3.1-flash-lite-preview-20260303", "runId": null, "seconds": null, "promptVersion": null }, "pageNumber": 18, "frameworkName": "authority-citation" } ], "arcBeats": [], "loops": [], "locked": true }, { "page": 19, "type": "appendix", "function": "appendix", "rawType": "appendix", "block": null, "metadata": { "slideType": "appendix", "slideTypeCanon": { "name": "Appendix", "slug": "appendix", "status": "active", "canonId": "019de52d-0762-71f8-b564-89ab863301cc", "version": 1, "description": null }, "function": "appendix", "functionCanon": { "name": "Appendix", "slug": "appendix", "status": "active", "canonId": "019de52d-138d-750b-83c3-49b4bc7a7a38", "version": 1, "description": null }, "density": "balanced", "densityScore": 35, "componentCount": 4, "textChars": 286, "nDataPoints": 0, "notes": null, "elementsJson": null, "confidence": 1, "extraction": { "model": "google/gemini-3.1-flash-lite-preview-20260303", "runId": null, "at": "2026-05-02 14:21:42.032+00", "seconds": 2.109, "promptVersion": "v2-canon-conditioned" } }, "links": { "slide": "/slides/019de06f-84e4-70e4-bce4-3c7c4f09a8ea/19", "deck": "/decks/019de06f-84e4-70e4-bce4-3c7c4f09a8ea", "deckAnchor": "/decks/019de06f-84e4-70e4-bce4-3c7c4f09a8ea#slide-19", "deckJson": "/decks/019de06f-84e4-70e4-bce4-3c7c4f09a8ea.json" }, "components": [ { "bbox": { "h": 0.2, "w": 0.8, "x": 0.05, "y": 0.2 }, "kind": "list", "text": "Co-authors on Learning from Human Preferences paper: Paul Christiano, Jan Leike, Tom Brown, Miljan Martik, Shane Legg", "attrs": null, "canon": { "kind": { "name": "List", "slug": "list", "status": "active", "canonId": "019de52c-f94b-779b-a53b-fd4904a5f4d2", "version": 1, "description": "Bullet or numbered list." }, "tool": null, "subkind": { "name": "Bullet list", "slug": "bullet", "status": "active", "canonId": "019de52c-fc98-7169-a083-5cab805076ca", "version": 1, "description": null }, "framework": null }, "subkind": "bullet", "groupRole": null, "confidence": null, "extraction": { "at": "2026-05-02 14:21:42.032+00", "model": "google/gemini-3.1-flash-lite-preview-20260303", "runId": null, "seconds": null, "promptVersion": null }, "componentId": "da46ac2d-0720-450a-97f3-5df16490fc1e", "parentComponentId": null }, { "bbox": { "h": 0.2, "w": 0.8, "x": 0.05, "y": 0.45 }, "kind": "list", "text": "Co-authors on Concrete Problems paper: Chris Olah, Jacob Steinhardt, Paul Christiano, John Schulman, Dan Mane", "attrs": null, "canon": { "kind": { "name": "List", "slug": "list", "status": "active", "canonId": "019de52c-f94b-779b-a53b-fd4904a5f4d2", "version": 1, "description": "Bullet or numbered list." }, "tool": null, "subkind": { "name": "Bullet list", "slug": "bullet", "status": "active", "canonId": "019de52c-fc98-7169-a083-5cab805076ca", "version": 1, "description": null }, "framework": null }, "subkind": "bullet", "groupRole": null, "confidence": null, "extraction": { "at": "2026-05-02 14:21:42.032+00", "model": "google/gemini-3.1-flash-lite-preview-20260303", "runId": null, "seconds": null, "promptVersion": null }, "componentId": "f287ef80-6b89-4608-bd00-e5cf09c67ec2", "parentComponentId": null }, { "bbox": { "h": 0.1, "w": 0.4, "x": 0.05, "y": 0.7 }, "kind": "list", "text": "Institutions: OpenAI, DeepMind, Google Brain", "attrs": null, "canon": { "kind": { "name": "List", "slug": "list", "status": "active", "canonId": "019de52c-f94b-779b-a53b-fd4904a5f4d2", "version": 1, "description": "Bullet or numbered list." }, "tool": null, "subkind": { "name": "Bullet list", "slug": "bullet", "status": "active", "canonId": "019de52c-fc98-7169-a083-5cab805076ca", "version": 1, "description": null }, "framework": null }, "subkind": "bullet", "groupRole": null, "confidence": null, "extraction": { "at": "2026-05-02 14:21:42.032+00", "model": "google/gemini-3.1-flash-lite-preview-20260303", "runId": null, "seconds": null, "promptVersion": null }, "componentId": "fb213b64-19c8-4b82-b1ad-c2e441b9017f", "parentComponentId": null }, { "bbox": { "h": 0.1, "w": 0.35, "x": 0.05, "y": 0.05 }, "kind": "title", "text": "Acknowledgements", "attrs": null, "canon": { "kind": { "name": "Title", "slug": "title", "status": "active", "canonId": "019de52c-f7dd-76d9-b848-b1f12bd85cf4", "version": 1, "description": "Slide title or headline." }, "tool": null, "subkind": { "name": "Headline", "slug": "headline", "status": "active", "canonId": "019de52c-fae1-7463-9a7c-d60668eced0d", "version": 1, "description": null }, "framework": null }, "subkind": "headline", "groupRole": null, "confidence": null, "extraction": { "at": "2026-05-02 14:21:42.032+00", "model": "google/gemini-3.1-flash-lite-preview-20260303", "runId": null, "seconds": null, "promptVersion": null }, "componentId": "a8054913-8bf7-4e7d-a9e1-56fa78c00945", "parentComponentId": null } ], "metrics": [], "tools": [], "frameworks": [], "arcBeats": [], "loops": [], "locked": true }, { "page": 20, "type": "setup", "function": "summarize", "rawType": "closing_ask", "block": null, "metadata": { "slideType": "closing_ask", "slideTypeCanon": { "name": "Closing ask", "slug": "closing_ask", "status": "active", "canonId": "019de52d-0aa4-750f-8aa1-8d622dd99f5c", "version": 1, "description": null }, "function": "summarize", "functionCanon": { "name": "Summarize", "slug": "summarize", "status": "active", "canonId": "019de52d-1541-704f-b86f-bf7b75056373", "version": 1, "description": null }, "density": "sparse", "densityScore": 0, "componentCount": 1, "textChars": 5, "nDataPoints": 0, "notes": null, "elementsJson": null, "confidence": 1, "extraction": { "model": "google/gemini-3.1-flash-lite-preview-20260303", "runId": null, "at": "2026-05-02 14:21:41.697+00", "seconds": 1.681, "promptVersion": "v2-canon-conditioned" } }, "links": { "slide": "/slides/019de06f-84e4-70e4-bce4-3c7c4f09a8ea/20", "deck": "/decks/019de06f-84e4-70e4-bce4-3c7c4f09a8ea", "deckAnchor": "/decks/019de06f-84e4-70e4-bce4-3c7c4f09a8ea#slide-20", "deckJson": "/decks/019de06f-84e4-70e4-bce4-3c7c4f09a8ea.json" }, "components": [ { "bbox": { "h": 0.16, "w": 0.28, "x": 0.36, "y": 0.42 }, "kind": "title", "text": "Q & A", "attrs": null, "canon": { "kind": { "name": "Title", "slug": "title", "status": "active", "canonId": "019de52c-f7dd-76d9-b848-b1f12bd85cf4", "version": 1, "description": "Slide title or headline." }, "tool": null, "subkind": { "name": "Headline", "slug": "headline", "status": "active", "canonId": "019de52c-fae1-7463-9a7c-d60668eced0d", "version": 1, "description": null }, "framework": null }, "subkind": "headline", "groupRole": null, "confidence": null, "extraction": { "at": "2026-05-02 14:21:41.697+00", "model": "google/gemini-3.1-flash-lite-preview-20260303", "runId": null, "seconds": null, "promptVersion": null }, "componentId": "421b27ff-ae70-4129-aebf-5c76393d2799", "parentComponentId": null } ], "metrics": [], "tools": [ { "tool": { "name": "Storytelling: the closing ask", "slug": "closing-ask", "status": "active", "bestFor": null, "canonId": "019df269-24e0-70ab-8dd9-983832c1c665", "version": 1, "bodyDocId": "019df269-24bc-7554-b88d-245127606b8d", "description": "The single most important slide in the deck — after the cover. The final", "familyLabel": "narrative-block", "categoryName": "Narrative & Transformation", "categorySlug": "narrative-transformation" }, "agent": "storyteller", "layer": "slide", "agents": [ "storyteller" ], "matchId": "e9be09aa-a02d-48c3-9424-2002a53ebf4e", "evidence": "The slide has a headline 'Q & A'.", "pageRefs": null, "priority": null, "whenToUse": null, "confidence": 0.8, "extraction": { "at": "2026-07-16 23:22:58.270719+00", "model": "or:meta-llama/llama-4-scout", "runId": null, "seconds": null, "promptVersion": null }, "pageNumber": 20, "whyItWorks": null, "antipattern": null, "cardinality": null, "narrativePurpose": null } ], "frameworks": [], "arcBeats": [], "loops": [], "locked": true } ], "arcBeats": [ { "from": 2, "to": 4, "label": "Problem", "description": "The deck identifies potential negative social impacts of AI and sources of accident risk." }, { "from": 5, "to": 6, "label": "Agitate", "description": "The deck highlights the risks of distributional shift and bad objective functions." }, { "from": 8, "to": 11, "label": "Solution", "description": "The deck proposes a solution using human feedback training processes and provides evidence to support it." } ], "loops": [ { "from": 2, "to": 6, "label": "Cost Of Inaction", "description": "What are the risks of not addressing AI safety?" }, { "from": 8, "to": 14, "label": "Jobs To Be Done", "description": "How can we train an ML system to align with human preferences and values?" } ] }