{
  "docId": "019dd923-5e88-73ef-bd5c-f812573a947a",
  "docSlug": "eea7524c557036f4",
  "documentTitle": "2020 Air Street Capital The State of AI Report 2020",
  "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": 17,
  "pageCount": 177,
  "prevPage": 16,
  "nextPage": 18,
  "slideType": "industry_trends",
  "function": "quantify_impact",
  "density": "overcrowded",
  "nDataPoints": 9,
  "notes": "The slide uses a mix of bulleted data points and narrative paragraphs to illustrate the cost barrier to entry for AI model training.",
  "elementsJson": [
    "headline_text",
    "action_title",
    "paragraph",
    "callout_box",
    "numbered_list"
  ],
  "metadataConfidence": 0.95,
  "imagePath": null,
  "slideHref": "/slides/019dd923-5e88-73ef-bd5c-f812573a947a/17",
  "deckHref": "/decks/019dd923-5e88-73ef-bd5c-f812573a947a",
  "deckJsonHref": "/decks/019dd923-5e88-73ef-bd5c-f812573a947a.json",
  "deckAnchorHref": "/decks/019dd923-5e88-73ef-bd5c-f812573a947a#slide-17",
  "components": [
    {
      "bbox": null,
      "kind": "callout",
      "text": "Based on variables released by Google et al., you're paying circa $1 per 1,000 parameters. This means OpenAI's 175B parameter GPT-3 could have cost tens of millions to train.",
      "attrs": null,
      "subkind": null,
      "toolName": "Visual emphasis",
      "toolSlug": "visual-emphasis",
      "confidence": null,
      "componentId": "019dd952-47c5-73ac-aa60-1f1848da78fc",
      "frameworkName": null,
      "frameworkSlug": null
    },
    {
      "bbox": {
        "h": 0.12,
        "w": 0.3,
        "x": 0.18,
        "y": 0.515
      },
      "kind": "list",
      "text": "$2.5k - $50k (110 million parameter model)\n$10k - $200k (340 million parameter model)\n$80k - $1.6m (1.5 billion parameter model)",
      "attrs": null,
      "subkind": "bullet",
      "toolName": null,
      "toolSlug": null,
      "confidence": null,
      "componentId": "ad60e401-5a11-40a7-90b6-fd58cc0b8552",
      "frameworkName": null,
      "frameworkSlug": null
    },
    {
      "bbox": null,
      "kind": "metric",
      "text": "Training cost: $10M",
      "attrs": null,
      "subkind": "primary",
      "toolName": "Quantification",
      "toolSlug": "quantification",
      "confidence": null,
      "componentId": "019dd952-47c5-73ac-aa60-203880288d74",
      "frameworkName": null,
      "frameworkSlug": null
    },
    {
      "bbox": {
        "h": 0.12,
        "w": 0.8,
        "x": 0.15,
        "y": 0.35
      },
      "kind": "paragraph",
      "text": "Just how much does it cost to train a model? Two correct answers are \"depends\" and \"a lot\". More quantitatively, here are current ballpark list-price costs of training differently sized BERT [4] models on the Wikipedia and Book corpora (15 GB).",
      "attrs": null,
      "subkind": "paragraph",
      "toolName": null,
      "toolSlug": null,
      "confidence": null,
      "componentId": "2df057fb-135b-4ba7-b040-6ce2d31db595",
      "frameworkName": null,
      "frameworkSlug": null
    },
    {
      "bbox": {
        "h": 0.1,
        "w": 0.8,
        "x": 0.15,
        "y": 0.78
      },
      "kind": "paragraph",
      "text": "Not many companies - certainly not many startups - can afford this cost. Some argue that this is not a severe issue; let the Googles of the world pre-train and publish the large language models, and let the rest of the world fine-tune them (a much cheaper endeavor) to specific tasks. Others (e.g., Etchemendy and Li [6]) are not as sanguine.",
      "attrs": null,
      "subkind": "paragraph",
      "toolName": null,
      "toolSlug": null,
      "confidence": null,
      "componentId": "4cb9c463-7bc5-4ea2-b168-989e987053fe",
      "frameworkName": null,
      "frameworkSlug": null
    },
    {
      "bbox": {
        "h": 0.12,
        "w": 0.7,
        "x": 0.15,
        "y": 0.66
      },
      "kind": "paragraph",
      "text": "For example, based on information released by Google, we estimate that, at list-price, training the 11B-parameter variant of T5 [5] cost well above $1.3 million for a single run. Assuming 2-3 runs of the large model and hundreds of the small ones, the (list-)price tag for the entire project may have been $10 million.",
      "attrs": null,
      "subkind": "paragraph",
      "toolName": null,
      "toolSlug": null,
      "confidence": null,
      "componentId": "e1ce7b1b-efdd-4a22-8835-0de21ba32cc9",
      "frameworkName": null,
      "frameworkSlug": null
    },
    {
      "bbox": {
        "h": 0.08,
        "w": 0.9,
        "x": 0.05,
        "y": 0.21
      },
      "kind": "paragraph",
      "text": "Based on variables released by Google et al., you're paying circa $1 per 1,000 parameters. This means OpenAI's 175B parameter GPT-3 could have cost tens of millions to train. Experts suggest the likely budget was $10M.",
      "attrs": null,
      "subkind": "paragraph",
      "toolName": null,
      "toolSlug": null,
      "confidence": null,
      "componentId": "fef45105-818d-4820-a992-a8d7933b6dc0",
      "frameworkName": null,
      "frameworkSlug": null
    },
    {
      "bbox": {
        "h": 0.05,
        "w": 0.65,
        "x": 0.05,
        "y": 0.13
      },
      "kind": "title",
      "text": "Tuning billions of model parameters costs millions of dollars",
      "attrs": null,
      "subkind": "headline",
      "toolName": null,
      "toolSlug": null,
      "confidence": null,
      "componentId": "cff22a54-40ad-4403-9a7d-dcc6f442e6bd",
      "frameworkName": null,
      "frameworkSlug": null
    }
  ],
  "metrics": [],
  "tools": [
    {
      "name": "Concrete Language",
      "slug": "concrete-language",
      "agent": "Storyteller",
      "layer": "slide",
      "matchId": "019dd95a-0fd5-7148-8ece-3fb0e01e620a",
      "evidence": "'$1 per 1,000 parameters' makes abstract cost concrete.",
      "confidence": 85
    },
    {
      "name": "Metaphor & Analogy",
      "slug": "metaphor-analogy",
      "agent": "Storyteller",
      "layer": "slide",
      "matchId": "019dd95a-0fd5-7148-8ece-416998989edf",
      "evidence": "Per-parameter pricing analogy to commodity goods.",
      "confidence": 60
    },
    {
      "name": "Quantification",
      "slug": "quantification",
      "agent": null,
      "layer": "slide",
      "matchId": "b4ef34b1-8767-4fa2-b07d-a118d5c11c81",
      "evidence": "The use of specific cost estimates ($1 per 1,000 parameters) and ranges ($2.5k - $50k) shows quantification of costs.",
      "confidence": 0.8
    }
  ],
  "frameworks": [],
  "arcBeats": [
    {
      "to": 129,
      "from": 10,
      "beatId": "019dd95a-0682-776c-8e34-ed2a8f38a754",
      "arcName": "The Triple Take",
      "arcSlug": "triple-take",
      "beatName": "The Facts (What)",
      "beatSlug": "triple-take-the-facts-what",
      "evidence": "Research, Talent, Industry sections inventory what happened in AI in 2020.",
      "position": 1,
      "confidence": 60,
      "parentBeatName": "Setup",
      "parentBeatSlug": "setup"
    },
    {
      "to": 129,
      "from": 10,
      "beatId": "019dd95a-0682-776c-8e34-fc384f439bba",
      "arcName": "The Mountain",
      "arcSlug": "mountain",
      "beatName": "Rising Action",
      "beatSlug": "mountain-rising-action",
      "evidence": "Research, Talent and Industry sections build momentum of AI progress.",
      "position": 2,
      "confidence": 40,
      "parentBeatName": "Development",
      "parentBeatSlug": "development"
    }
  ],
  "loops": [
    {
      "to": 23,
      "from": 15,
      "name": "Pattern Hunter",
      "slug": "02-pattern-hunter",
      "bestFor": "Time-pressed audiences, building consensus, when data is strong",
      "matchId": "019dd95a-07fe-70ce-8d3c-ee7c2907ccd3",
      "evidence": "Slides 15-19 stack scaling-law evidence; 20-23 add caveats.",
      "position": 2,
      "objective": "Establish pattern that bigger models + more compute drive performance at rising cost",
      "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
}