{
  "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": 35,
  "pageCount": 163,
  "prevPage": 34,
  "nextPage": 36,
  "slideType": "industry_trends",
  "function": "present_framework",
  "density": "dense",
  "nDataPoints": 3,
  "notes": "The slide compares different prompting architectures (IO, CoT, CoT-SC, ToT) visually.",
  "elementsJson": [
    "headline_text",
    "bullet_list",
    "paragraph",
    "process_diagram"
  ],
  "metadataConfidence": 0.95,
  "imagePath": null,
  "slideHref": "/slides/019dd923-5e88-73ef-bd5d-06b04d219fea/35",
  "deckHref": "/decks/019dd923-5e88-73ef-bd5d-06b04d219fea",
  "deckJsonHref": "/decks/019dd923-5e88-73ef-bd5d-06b04d219fea.json",
  "deckAnchorHref": "/decks/019dd923-5e88-73ef-bd5d-06b04d219fea#slide-35",
  "components": [
    {
      "bbox": null,
      "kind": "callout",
      "text": "Optimization by Prompting (OPRO) shows that optimized prompts outperform human-designed prompts on GSM8K and Big-Bench Hard by a significant margin, sometimes over 50%.",
      "attrs": null,
      "subkind": null,
      "toolName": "Visual emphasis",
      "toolSlug": "visual-emphasis",
      "confidence": null,
      "componentId": "019dd952-47c4-7719-a227-1cfb088bfab4",
      "frameworkName": null,
      "frameworkSlug": null
    },
    {
      "bbox": {
        "h": 0.35,
        "w": 0.4,
        "x": 0.55,
        "y": 0.55
      },
      "kind": "image",
      "text": "Diagrams of IO, CoT, CoT-SC, and ToT prompting architectures.",
      "attrs": null,
      "subkind": "infographic",
      "toolName": null,
      "toolSlug": null,
      "confidence": null,
      "componentId": "fb84cb91-f3b0-46cf-acd4-3c0dd9666b2a",
      "frameworkName": null,
      "frameworkSlug": null
    },
    {
      "bbox": {
        "h": 0.35,
        "w": 0.5,
        "x": 0.02,
        "y": 0.54
      },
      "kind": "list",
      "text": "It turns out that LLMs are also great prompt engineers. Auto-CoT matches or exceeds the performance of CoT on 10 reasoning tasks. Automatic Prompt Engineer (APE) shows the same on 19/24 tasks. APE-engineered prompts are also able to steer models towards truthfulness and/or informativeness. Optimization by Prompting (OPRO) shows that optimized prompts outperform human-designed prompts on GSM8K and Big-Bench Hard by a significant margin, sometimes over 50%.",
      "attrs": null,
      "subkind": "bullet",
      "toolName": null,
      "toolSlug": null,
      "confidence": null,
      "componentId": "42ff9541-0f87-44d6-bd1f-a5c90733e761",
      "frameworkName": null,
      "frameworkSlug": null
    },
    {
      "bbox": {
        "h": 0.15,
        "w": 0.95,
        "x": 0.02,
        "y": 0.22
      },
      "kind": "list",
      "text": "The quality of a prompt highly influences task performance. Chain of Thought prompting (CoT) asks the LLM to additionally output intermediate reasoning steps which gives a boost in performance. Tree of Thought (ToT) further improves on that by sampling multiple times and representing the “thoughts” as nodes in a tree structure.",
      "attrs": null,
      "subkind": "bullet",
      "toolName": null,
      "toolSlug": null,
      "confidence": null,
      "componentId": "8e4c7849-11f5-49ff-ac76-096ffa22ad44",
      "frameworkName": null,
      "frameworkSlug": null
    },
    {
      "bbox": {
        "h": 0.15,
        "w": 0.95,
        "x": 0.02,
        "y": 0.38
      },
      "kind": "list",
      "text": "The tree structure of a ToT can be explored with a variety of search algorithms. In order to leverage this search, the LLM also needs to assign a value to node, for instance by classifying it as one of sure, likely or impossible. Graph of Thought (GoT) turns this reasoning tree into a graph by combining similar nodes.",
      "attrs": null,
      "subkind": "bullet",
      "toolName": null,
      "toolSlug": null,
      "confidence": null,
      "componentId": "9c75f7ba-b524-40fc-84f6-c4076e81a5f3",
      "frameworkName": null,
      "frameworkSlug": null
    },
    {
      "bbox": {
        "h": 0.05,
        "w": 0.75,
        "x": 0.02,
        "y": 0.14
      },
      "kind": "title",
      "text": "Where are we prompting? Take a deep breath...it's getting sophisticated",
      "attrs": null,
      "subkind": "headline",
      "toolName": null,
      "toolSlug": null,
      "confidence": null,
      "componentId": "7f9be260-59da-4740-9737-ad8beed96dcb",
      "frameworkName": null,
      "frameworkSlug": null
    }
  ],
  "metrics": [],
  "tools": [
    {
      "name": "Action Titles",
      "slug": "action-titles",
      "agent": "Architect",
      "layer": "slide",
      "matchId": "019dd95a-1055-74e0-a957-45288aac3f4e",
      "evidence": "Title quotes a meme prompt 'Take a deep breath...' as evidence of sophistication.",
      "confidence": 75
    },
    {
      "name": "Analytical method",
      "slug": "analytical-method",
      "agent": null,
      "layer": "slide",
      "matchId": "29156ab9-5450-4a3e-a8f3-38329da447cf",
      "evidence": "The quality of a prompt highly influences task performance.",
      "confidence": 0.6
    }
  ],
  "frameworks": [
    {
      "name": "Chain of Thought",
      "slug": null,
      "matchId": "da3e5d2c-e07c-4047-8ea8-03bd18cf4c01",
      "evidence": "Text mentions CoT, ToT, GoT, APE, OPRO",
      "confidence": 1
    }
  ],
  "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
}