{
  "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": 40,
  "pageCount": 163,
  "prevPage": 39,
  "nextPage": 41,
  "slideType": "case_study",
  "function": "illustrate_case",
  "density": "dense",
  "nDataPoints": 27,
  "notes": "Includes a diagram of the reasoning module (DAG) and a performance comparison table.",
  "elementsJson": [
    "headline_text",
    "paragraph",
    "bullet_list",
    "process_diagram",
    "data_table"
  ],
  "metadataConfidence": 1,
  "imagePath": null,
  "slideHref": "/slides/019dd923-5e88-73ef-bd5d-06b04d219fea/40",
  "deckHref": "/decks/019dd923-5e88-73ef-bd5d-06b04d219fea",
  "deckJsonHref": "/decks/019dd923-5e88-73ef-bd5d-06b04d219fea.json",
  "deckAnchorHref": "/decks/019dd923-5e88-73ef-bd5d-06b04d219fea#slide-40",
  "components": [
    {
      "bbox": null,
      "kind": "callout",
      "text": "It outperforms state-of-the-art RL baselines in open-world games with no training.",
      "attrs": null,
      "subkind": null,
      "toolName": "Visual emphasis",
      "toolSlug": "visual-emphasis",
      "confidence": null,
      "componentId": "019dd952-47c4-7719-a223-8dfccfd88b96",
      "frameworkName": null,
      "frameworkSlug": null
    },
    {
      "bbox": {
        "h": 0.35,
        "w": 0.45,
        "x": 0.03,
        "y": 0.55
      },
      "kind": "diagram",
      "text": "Reasoning module diagram showing observation processing and DAG-based reasoning",
      "attrs": null,
      "subkind": "process",
      "toolName": null,
      "toolSlug": null,
      "confidence": null,
      "componentId": "b2f346c2-9aea-4465-acba-50509879b79d",
      "frameworkName": null,
      "frameworkSlug": null
    },
    {
      "bbox": {
        "h": 0.12,
        "w": 0.95,
        "x": 0.03,
        "y": 0.31
      },
      "kind": "list",
      "text": "RL has been the go-to for game-based problems like Minecraft and Crafter, despite it being limited by the high sample complexity and difficulty in incorporating prior knowledge. In contrast, the LLM can processes the latex source of the paper and reasons through a QA framework (directed acyclic graph with questions as nodes and dependencies as edges) to take an environment action.",
      "attrs": null,
      "subkind": "bullet",
      "toolName": null,
      "toolSlug": null,
      "confidence": null,
      "componentId": "29ce3d50-982b-4129-8d06-395894606b26",
      "frameworkName": null,
      "frameworkSlug": null
    },
    {
      "bbox": {
        "h": 0.08,
        "w": 0.95,
        "x": 0.03,
        "y": 0.21
      },
      "kind": "list",
      "text": "Another text-only agent based on GPT-4 is SPRING. It outperforms state-of-the-art RL baselines in open-world games with no training. It reads a game's original academic paper and plays the game through an LLM.",
      "attrs": null,
      "subkind": "bullet",
      "toolName": null,
      "toolSlug": null,
      "confidence": null,
      "componentId": "e3be0e3f-3596-4e0c-a24e-277403d61c3d",
      "frameworkName": null,
      "frameworkSlug": null
    },
    {
      "bbox": null,
      "kind": "metric",
      "text": "Reward: 12.3",
      "attrs": null,
      "subkind": "primary",
      "toolName": "Quantification",
      "toolSlug": "quantification",
      "confidence": null,
      "componentId": "019dd952-47c4-7719-a223-9373b286042f",
      "frameworkName": null,
      "frameworkSlug": null
    },
    {
      "bbox": {
        "h": 0.35,
        "w": 0.45,
        "x": 0.5,
        "y": 0.55
      },
      "kind": "table",
      "text": "Performance comparison table of SPRING vs RL baselines",
      "attrs": null,
      "subkind": "data",
      "toolName": null,
      "toolSlug": null,
      "confidence": null,
      "componentId": "dd36e3f9-7e30-4063-bedf-27af3926dccf",
      "frameworkName": null,
      "frameworkSlug": null
    },
    {
      "bbox": {
        "h": 0.04,
        "w": 0.6,
        "x": 0.03,
        "y": 0.14
      },
      "kind": "title",
      "text": "GPT-4 out-performs RL algorithms by studying papers and reasoning",
      "attrs": null,
      "subkind": "headline",
      "toolName": null,
      "toolSlug": null,
      "confidence": null,
      "componentId": "6f3a4cd9-d385-4e5f-b9e7-25f8a4120433",
      "frameworkName": null,
      "frameworkSlug": null
    }
  ],
  "metrics": [],
  "tools": [
    {
      "name": "Comparison frame",
      "slug": "comparison-frame",
      "agent": null,
      "layer": "slide",
      "matchId": "13c9cc48-3399-4b54-bd57-aadd40e4f9a8",
      "evidence": "Performance comparison table of SPRING vs RL baselines",
      "confidence": 0.7
    }
  ],
  "frameworks": [
    {
      "name": "causal-chain",
      "slug": null,
      "matchId": "81ad8a25-59a4-45f2-926f-bfddb981de70",
      "evidence": "The reasoning module uses a directed acyclic graph to map dependencies between questions and actions.",
      "confidence": 0.7
    }
  ],
  "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": 40,
      "from": 32,
      "name": "Pattern Hunter",
      "slug": "02-pattern-hunter",
      "bestFor": "Time-pressed audiences, building consensus, when data is strong",
      "matchId": "019dd95a-07fe-70ce-8d3d-ec11d4feb687",
      "evidence": "Vibe-check, code, prompting, Agent Smith, Voyager, RAP, GPT-4-RL slides accumulate into 'LLMs as agents'.",
      "position": 4,
      "objective": "Show LLMs becoming agentic across prompting, tools and reasoning",
      "structure": "Evidence A -> Evidence B -> Evidence C -> Pattern/Conclusion",
      "confidence": 70,
      "description": "Group multiple pieces of evidence that together point to a pattern or conclusion"
    }
  ],
  "imagePathAlt": null,
  "thumbSrc": null,
  "thumbSrcAlt": null,
  "locked": true
}