{
  "docId": "019dd923-5e88-73ef-bd5d-06b04d219fea",
  "docSlug": "dd91c78f6570bf29",
  "documentTitle": "2023 Air Street Capital The State of AI Report 2023",
  "authorId": "AirStreetCapital",
  "authorName": "Air Street Capital",
  "documentKindSlug": "consulting-deck",
  "documentKindLabel": "Consulting deck",
  "sourceTypeSlug": "vc_research",
  "sourceTypeLabel": "VC research",
  "presentationDate": null,
  "orientation": "landscape",
  "aspectRatio": 1.777,
  "pageNumber": 49,
  "pageCount": 163,
  "prevPage": 48,
  "nextPage": 50,
  "slideType": "case_study",
  "function": "illustrate_case",
  "density": "dense",
  "nDataPoints": 2,
  "notes": "The slide explains the technical architecture of CICERO, including its use of dialogue history, board state, and a 2.7B-parameter BART-like model.",
  "elementsJson": [
    "headline_text",
    "bullet_list",
    "process_diagram"
  ],
  "metadataConfidence": 1,
  "imagePath": null,
  "slideHref": "/slides/019dd923-5e88-73ef-bd5d-06b04d219fea/49",
  "deckHref": "/decks/019dd923-5e88-73ef-bd5d-06b04d219fea",
  "deckJsonHref": "/decks/019dd923-5e88-73ef-bd5d-06b04d219fea.json",
  "deckAnchorHref": "/decks/019dd923-5e88-73ef-bd5d-06b04d219fea#slide-49",
  "components": [
    {
      "bbox": null,
      "kind": "callout",
      "text": "CICERO achieved double the average score of human players online and ranked in the top 10% players who played more than one game.",
      "attrs": null,
      "subkind": null,
      "toolName": "Visual emphasis",
      "toolSlug": "visual-emphasis",
      "confidence": null,
      "componentId": "019dd952-47c4-7719-a224-c02a20d24a7b",
      "frameworkName": null,
      "frameworkSlug": null
    },
    {
      "bbox": {
        "h": 0.55,
        "w": 0.22,
        "x": 0.75,
        "y": 0.35
      },
      "kind": "diagram",
      "text": "Technical architecture diagram showing inputs (Board State, Dialogue History) flowing into a Dialogue-Conditional Action Model, Simulator, and Dialogue model.",
      "attrs": null,
      "subkind": "process",
      "toolName": null,
      "toolSlug": null,
      "confidence": null,
      "componentId": "193483d1-ddd0-4a04-b7c5-7f5718b87000",
      "frameworkName": null,
      "frameworkSlug": null
    },
    {
      "bbox": {
        "h": 0.5,
        "w": 0.65,
        "x": 0.03,
        "y": 0.38
      },
      "kind": "list",
      "text": "Fast parallel progress in strategic planning and language modeling allows for potentially great advancements at the intersection, with applications in human-AI cooperation. Meta tackles the game of Diplomacy as a benchmark for such progress.\nCICERO uses dialogue history between players as well as the board state and its history to begin predicting what everyone will do. It then iteratively refines these predictions using planning, then decides according to a policy which action it intends to take. CICERO then generates and filters candidate messages to communicate with players.\nThe controllable dialogue model it uses is based on a 2.7B-params BART-like model fine-tuned on >40K online games of Diplomacy. CICERO uses a new iterative planning algorithm based on piKL which improves the predictions of other players' moves after dialoguing with them.",
      "attrs": null,
      "subkind": "bullet",
      "toolName": null,
      "toolSlug": null,
      "confidence": null,
      "componentId": "44624b58-6fed-41b9-b6ab-c4e447b83325",
      "frameworkName": null,
      "frameworkSlug": null
    },
    {
      "bbox": null,
      "kind": "metric",
      "text": "player ranking: 10%",
      "attrs": null,
      "subkind": "primary",
      "toolName": "Quantification",
      "toolSlug": "quantification",
      "confidence": null,
      "componentId": "019dd952-47c4-7719-a224-c7c388abeefd",
      "frameworkName": null,
      "frameworkSlug": null
    },
    {
      "bbox": {
        "h": 0.12,
        "w": 0.65,
        "x": 0.03,
        "y": 0.22
      },
      "kind": "paragraph",
      "text": "Meta trained an AI agent to play a popular multiplayer strategy game called Diplomacy, which involves planning and negotiating in natural language with other players over multiple rounds. CICERO achieved double the average score of human players online and ranked in the top 10% players who played more than one game.",
      "attrs": null,
      "subkind": "paragraph",
      "toolName": null,
      "toolSlug": null,
      "confidence": null,
      "componentId": "acc0b9cd-a8fb-49f1-95bf-1f02e0895473",
      "frameworkName": null,
      "frameworkSlug": null
    },
    {
      "bbox": {
        "h": 0.05,
        "w": 0.6,
        "x": 0.03,
        "y": 0.14
      },
      "kind": "title",
      "text": "CICERO masters natural language to beat humans at Diplomacy",
      "attrs": null,
      "subkind": "headline",
      "toolName": null,
      "toolSlug": null,
      "confidence": null,
      "componentId": "96a5ec18-f13c-4c9e-8808-5c815d202122",
      "frameworkName": null,
      "frameworkSlug": null
    }
  ],
  "metrics": [],
  "tools": [
    {
      "name": "Sinatra Test",
      "slug": "sinatra-test",
      "agent": "Storyteller",
      "layer": "slide",
      "matchId": "019dd95a-1055-74e0-a957-5632eba587bd",
      "evidence": "CICERO beating humans at Diplomacy is a flagship-example case for NL+strategy.",
      "confidence": 70
    }
  ],
  "frameworks": [],
  "arcBeats": [
    {
      "to": 120,
      "from": 11,
      "beatId": "019dd95a-0682-776c-8e35-41afd44ef59f",
      "arcName": "The Triple Take",
      "arcSlug": "triple-take",
      "beatName": "The Facts (What)",
      "beatSlug": "triple-take-the-facts-what",
      "evidence": "Research + Industry sections inventory model, compute, funding facts.",
      "position": 1,
      "confidence": 78,
      "parentBeatName": "Setup",
      "parentBeatSlug": "setup"
    },
    {
      "to": 120,
      "from": 11,
      "beatId": "019dd95a-0682-776c-8e35-523bfb7f96e6",
      "arcName": "The Mountain",
      "arcSlug": "mountain",
      "beatName": "Rising Action",
      "beatSlug": "mountain-rising-action",
      "evidence": "Escalating capabilities, compute concentration and capital flows.",
      "position": 2,
      "confidence": 45,
      "parentBeatName": "Development",
      "parentBeatSlug": "development"
    }
  ],
  "loops": [
    {
      "to": 49,
      "from": 41,
      "name": "Pattern Hunter",
      "slug": "02-pattern-hunter",
      "bestFor": "Time-pressed audiences, building consensus, when data is strong",
      "matchId": "019dd95a-07fe-70ce-8d3d-f0e6f3668fb1",
      "evidence": "VLM, autonomous driving, PaLM-E, RT-2, RoboCat, drone-racing, CICERO stack as parallel evidence.",
      "position": 5,
      "objective": "Multimodal models extend from vision-language to embodied robots",
      "structure": "Evidence A -> Evidence B -> Evidence C -> Pattern/Conclusion",
      "confidence": 75,
      "description": "Group multiple pieces of evidence that together point to a pattern or conclusion"
    }
  ],
  "imagePathAlt": null,
  "thumbSrc": null,
  "thumbSrcAlt": null,
  "locked": true
}