{
  "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": 19,
  "pageCount": 163,
  "prevPage": 18,
  "nextPage": 20,
  "slideType": "case_study",
  "function": "illustrate_case",
  "density": "dense",
  "nDataPoints": 18,
  "notes": "The slide uses a stacked bar chart to show human evaluation of LLaMa-2 helpfulness against other models.",
  "elementsJson": [
    "headline_text",
    "action_title",
    "bullet_list",
    "bar_chart_stacked"
  ],
  "metadataConfidence": 1,
  "imagePath": null,
  "slideHref": "/slides/019dd923-5e88-73ef-bd5d-06b04d219fea/19",
  "deckHref": "/decks/019dd923-5e88-73ef-bd5d-06b04d219fea",
  "deckJsonHref": "/decks/019dd923-5e88-73ef-bd5d-06b04d219fea.json",
  "deckAnchorHref": "/decks/019dd923-5e88-73ef-bd5d-06b04d219fea#slide-19",
  "components": [
    {
      "bbox": null,
      "kind": "callout",
      "text": "LLaMa-2 70B is competitive with ChatGPT on most tasks except for coding, where it significantly lags behind it.",
      "attrs": null,
      "subkind": null,
      "toolName": "Visual emphasis",
      "toolSlug": "visual-emphasis",
      "confidence": null,
      "componentId": "019dd952-47c4-7719-a221-4d007c4b945d",
      "frameworkName": null,
      "frameworkSlug": null
    },
    {
      "bbox": {
        "h": 0.35,
        "w": 0.32,
        "x": 0.65,
        "y": 0.45
      },
      "kind": "chart",
      "text": "Human evaluation of LLaMa-2 helpfulness vs. other open source models",
      "attrs": null,
      "subkind": "bar-stacked-100pct",
      "toolName": null,
      "toolSlug": null,
      "confidence": null,
      "componentId": "46a044e1-39a4-4c9b-8ce8-94fc3509efe7",
      "frameworkName": null,
      "frameworkSlug": null
    },
    {
      "bbox": {
        "h": 0.7,
        "w": 0.6,
        "x": 0.02,
        "y": 0.22
      },
      "kind": "list",
      "text": "In July'23, the LLaMa-2 series of models was released, giving (almost) everyone the right for commercial use. The base LLaMa-2 model is almost identical to LLaMa-1 but further fine-tuned using instruction tuning and RLHF and optimized for dialogue applications. In September 2023, Llama-2 as had almost 32M downloads.\nThe pre-training corpus for LLaMa-2 has 2 trillion tokens (40% increase).\nFor supervised fine-tuning, the researchers tried publicly available data, but what was most helpful was using a few (24,540) high-quality vendor-based annotations. For RLHF, they use binary comparison and split the RLHF process into prompts and answers designed to be helpful to the user and others designed to be safe.\nLLaMa-2 70B is competitive with ChatGPT on most tasks except for coding, where it significantly lags behind it. But CodeLLaMa, a fine-tuned version for code beats all non-GPT4 models (more on this later).\nPer Meta terms, anyone (with enough hardware to run the models) can use the LLaMa-2 models, as long as their commercial application didn't have more than 700M users at the time of LLaMa-2's release.",
      "attrs": null,
      "subkind": "bullet",
      "toolName": null,
      "toolSlug": null,
      "confidence": null,
      "componentId": "d5374fd3-9b3d-495e-81e2-b49edcc54cf8",
      "frameworkName": null,
      "frameworkSlug": null
    },
    {
      "bbox": null,
      "kind": "metric",
      "text": "Human evaluation of helpfulness: 32M",
      "attrs": null,
      "subkind": "primary",
      "toolName": "Quantification",
      "toolSlug": "quantification",
      "confidence": null,
      "componentId": "019dd952-47c4-7719-a221-53ad840403f1",
      "frameworkName": null,
      "frameworkSlug": null
    },
    {
      "bbox": {
        "h": 0.05,
        "w": 0.6,
        "x": 0.02,
        "y": 0.14
      },
      "kind": "title",
      "text": "LLaMa-2: the most generally capable and publicly accessible LLM?",
      "attrs": null,
      "subkind": "headline",
      "toolName": null,
      "toolSlug": null,
      "confidence": null,
      "componentId": "c224e14b-60d7-40f7-b66e-8cc380c2ddc4",
      "frameworkName": null,
      "frameworkSlug": null
    }
  ],
  "metrics": [],
  "tools": [
    {
      "name": "Concrete Language",
      "slug": "concrete-language",
      "agent": "Storyteller",
      "layer": "slide",
      "matchId": "019dd95a-1055-74e0-a957-1f68259fda5c",
      "evidence": "'LLaMa-2 70B is competitive with ChatGPT on most tasks except for coding' is a specific testable claim.",
      "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": 21,
      "from": 12,
      "name": "Pattern Hunter",
      "slug": "02-pattern-hunter",
      "bestFor": "Time-pressed audiences, building consensus, when data is strong",
      "matchId": "019dd95a-07fe-70ce-8d3d-e61d00ffc6e4",
      "evidence": "Sequential GPT-4, RLHF, LLaMa, popularity-mention slides converge on a single conclusion.",
      "position": 2,
      "objective": "Pile up evidence that GPT-4 and RLHF dominate the LLM landscape",
      "structure": "Evidence A -> Evidence B -> Evidence C -> Pattern/Conclusion",
      "confidence": 82,
      "description": "Group multiple pieces of evidence that together point to a pattern or conclusion"
    }
  ],
  "imagePathAlt": null,
  "thumbSrc": null,
  "thumbSrcAlt": null,
  "locked": true
}