{
  "docId": "019dd923-5e88-73ef-bd5d-0856e1444fb9",
  "docSlug": "8757f1b44ef7f176",
  "documentTitle": "2024 Air Street Capital The State of AI Report 2024",
  "authorId": "AirStreetCapital",
  "authorName": "Air Street Capital",
  "documentKindSlug": "consulting-deck",
  "documentKindLabel": "Consulting deck",
  "sourceTypeSlug": "vc_research",
  "sourceTypeLabel": "VC research",
  "presentationDate": null,
  "orientation": "landscape",
  "aspectRatio": 1.777,
  "pageNumber": 78,
  "pageCount": 213,
  "prevPage": 77,
  "nextPage": 79,
  "slideType": "industry_trends",
  "function": "summarize",
  "density": "overcrowded",
  "nDataPoints": 0,
  "notes": "The slide highlights two specific research directions: learning affordance from human video (CMU) and chain-of-thought reasoning (Berkeley/Stanford).",
  "elementsJson": [
    "headline_text",
    "bullet_list",
    "paragraph",
    "process_diagram",
    "icon_grid"
  ],
  "metadataConfidence": 0.95,
  "imagePath": null,
  "slideHref": "/slides/019dd923-5e88-73ef-bd5d-0856e1444fb9/78",
  "deckHref": "/decks/019dd923-5e88-73ef-bd5d-0856e1444fb9",
  "deckJsonHref": "/decks/019dd923-5e88-73ef-bd5d-0856e1444fb9.json",
  "deckAnchorHref": "/decks/019dd923-5e88-73ef-bd5d-0856e1444fb9#slide-78",
  "components": [
    {
      "bbox": null,
      "kind": "callout",
      "text": "Rather than finding more data, researchers are injecting more structure and knowledge to what we already have.",
      "attrs": null,
      "subkind": null,
      "toolName": "Visual emphasis",
      "toolSlug": "visual-emphasis",
      "confidence": null,
      "componentId": "019dd952-47c3-7407-8d15-39358b41250e",
      "frameworkName": null,
      "frameworkSlug": null
    },
    {
      "bbox": {
        "h": 0.48,
        "w": 0.27,
        "x": 0.71,
        "y": 0.41
      },
      "kind": "image",
      "text": "Comparison diagram of Regular VLA Policy vs VLA w/ Embodied Chain-of-Thought",
      "attrs": null,
      "subkind": "infographic",
      "toolName": null,
      "toolSlug": null,
      "confidence": null,
      "componentId": "4e1d4095-1670-4822-b83a-6e67372736cf",
      "frameworkName": null,
      "frameworkSlug": null
    },
    {
      "bbox": {
        "h": 0.12,
        "w": 0.68,
        "x": 0.02,
        "y": 0.34
      },
      "kind": "list",
      "text": "One approach, outlined by a Carnegie Mellon team, involves learning more “affordance” information from human video data, such as hand possess, object interactions, and contact points.",
      "attrs": null,
      "subkind": "bullet",
      "toolName": null,
      "toolSlug": null,
      "confidence": null,
      "componentId": "15c69acf-b032-452b-a085-e7bad24ec891",
      "frameworkName": null,
      "frameworkSlug": null
    },
    {
      "bbox": {
        "h": 0.12,
        "w": 0.68,
        "x": 0.02,
        "y": 0.64
      },
      "kind": "list",
      "text": "Rather than just predicting actions directly, the enhanced models are trained to reason step-by-step about plans, sub-tasks, and visual features before deciding on actions.",
      "attrs": null,
      "subkind": "bullet",
      "toolName": null,
      "toolSlug": null,
      "confidence": null,
      "componentId": "5f6733e6-93b4-45d2-8eb2-2b1f51f20465",
      "frameworkName": null,
      "frameworkSlug": null
    },
    {
      "bbox": {
        "h": 0.12,
        "w": 0.68,
        "x": 0.02,
        "y": 0.22
      },
      "kind": "list",
      "text": "Robotics policies have often been hampered by a lack of generalizability, due to limited real-world data. Rather than finding more data, researchers are injecting more structure and knowledge to what we already have.",
      "attrs": null,
      "subkind": "bullet",
      "toolName": null,
      "toolSlug": null,
      "confidence": null,
      "componentId": "c88eb924-9aaa-4958-9ddf-22c97a928b43",
      "frameworkName": null,
      "frameworkSlug": null
    },
    {
      "bbox": {
        "h": 0.12,
        "w": 0.68,
        "x": 0.02,
        "y": 0.46
      },
      "kind": "list",
      "text": "This information can then be used to finetine existing visual representations to make them more for suitable robotic tasks. This consistently improved performance on real-world manipulation tasks.",
      "attrs": null,
      "subkind": "bullet",
      "toolName": null,
      "toolSlug": null,
      "confidence": null,
      "componentId": "d1c8dfcf-bf3a-4f40-833d-1de6b44badaf",
      "frameworkName": null,
      "frameworkSlug": null
    },
    {
      "bbox": {
        "h": 0.06,
        "w": 0.68,
        "x": 0.02,
        "y": 0.58
      },
      "kind": "list",
      "text": "Meanwhile, a Berkeley/Stanford team found that chain-of-thought reasoning could have a similar impact.",
      "attrs": null,
      "subkind": "bullet",
      "toolName": null,
      "toolSlug": null,
      "confidence": null,
      "componentId": "eaaef8f1-2c5d-45c5-afed-611b3583bda7",
      "frameworkName": null,
      "frameworkSlug": null
    },
    {
      "bbox": {
        "h": 0.06,
        "w": 0.68,
        "x": 0.02,
        "y": 0.76
      },
      "kind": "list",
      "text": "This approach uses LLMs to generate training data for the reasoning steps.",
      "attrs": null,
      "subkind": "bullet",
      "toolName": null,
      "toolSlug": null,
      "confidence": null,
      "componentId": "fb2e1fde-88b7-45ad-9a3e-e18dc4612a6e",
      "frameworkName": null,
      "frameworkSlug": null
    },
    {
      "bbox": {
        "h": 0.03,
        "w": 0.6,
        "x": 0.02,
        "y": 0.03
      },
      "kind": "other",
      "text": "Introduction | Research | Industry | Politics | Safety | Predictions",
      "attrs": null,
      "subkind": "unclassified",
      "toolName": null,
      "toolSlug": null,
      "confidence": null,
      "componentId": "2fd7604d-0393-42f0-b593-e01eb56f4cfb",
      "frameworkName": null,
      "frameworkSlug": null
    },
    {
      "bbox": {
        "h": 0.05,
        "w": 0.85,
        "x": 0.02,
        "y": 0.13
      },
      "kind": "title",
      "text": "Can we stretch existing real-world robotics data further than we currently do?",
      "attrs": null,
      "subkind": "headline",
      "toolName": null,
      "toolSlug": null,
      "confidence": null,
      "componentId": "97b2ce17-29be-4b4f-bd6b-b701fd6e5301",
      "frameworkName": null,
      "frameworkSlug": null
    }
  ],
  "metrics": [],
  "tools": [
    {
      "name": "Problem Statement Canvas",
      "slug": "problem-statement-canvas",
      "agent": "Architect",
      "layer": "slide",
      "matchId": "9510f602-f22a-4a28-9def-7d37b54a3f10",
      "evidence": "the components discuss challenges and approaches to overcoming data bottlenecks in robotics",
      "confidence": 0.6
    }
  ],
  "frameworks": [
    {
      "name": "chain-of-thought",
      "slug": null,
      "matchId": "9676c843-2587-4e24-b7a2-f8ed1078531f",
      "evidence": "Explicit mention of 'chain-of-thought reasoning' in the text and diagram.",
      "confidence": 1
    }
  ],
  "arcBeats": [
    {
      "to": 153,
      "from": 9,
      "beatId": "019dd95a-0682-776c-8e35-5f2398b8d1d0",
      "arcName": "The Triple Take",
      "arcSlug": "triple-take",
      "beatName": "The Facts (What)",
      "beatSlug": "triple-take-the-facts-what",
      "evidence": "Exec summary + Research + Industry sections inventory the year",
      "position": 1,
      "confidence": 78,
      "parentBeatName": "Setup",
      "parentBeatSlug": "setup"
    },
    {
      "to": 85,
      "from": 12,
      "beatId": "019dd95a-0682-776c-8e35-6d9d8ee06d0a",
      "arcName": "Voyage and Return",
      "arcSlug": "voyage-return",
      "beatName": "The Unknown",
      "beatSlug": "voyage-return-the-unknown",
      "evidence": "Research section explores frontier model uncharted territory",
      "position": 2,
      "confidence": 55,
      "parentBeatName": "Complication",
      "parentBeatSlug": "complication"
    }
  ],
  "loops": [
    {
      "to": 81,
      "from": 72,
      "name": "Pattern Hunter",
      "slug": "02-pattern-hunter",
      "bestFor": "Time-pressed audiences, building consensus, when data is strong",
      "matchId": "019dd95a-07fe-70ce-8d3e-431790a1bfd7",
      "evidence": "Wayve LINGO-2, SAM2, DeepMind robotics, LeRobot, diffusion policies, humanoids, Vision Pro",
      "position": 9,
      "objective": "Show robotics renaissance via converging evidence",
      "structure": "Evidence A -> Evidence B -> Evidence C -> Pattern/Conclusion",
      "confidence": 78,
      "description": "Group multiple pieces of evidence that together point to a pattern or conclusion"
    }
  ],
  "imagePathAlt": null,
  "thumbSrc": null,
  "thumbSrcAlt": null,
  "locked": true
}