{
  "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": 99,
  "pageCount": 213,
  "prevPage": 98,
  "nextPage": 100,
  "slideType": "case_study",
  "function": "illustrate_case",
  "density": "dense",
  "nDataPoints": 15,
  "notes": "The slide uses a pie chart to visualize the root causes of infrastructure failures in large-scale AI training.",
  "elementsJson": [
    "headline_text",
    "paragraph",
    "pie_chart"
  ],
  "metadataConfidence": 1,
  "imagePath": null,
  "slideHref": "/slides/019dd923-5e88-73ef-bd5d-0856e1444fb9/99",
  "deckHref": "/decks/019dd923-5e88-73ef-bd5d-0856e1444fb9",
  "deckJsonHref": "/decks/019dd923-5e88-73ef-bd5d-0856e1444fb9.json",
  "deckAnchorHref": "/decks/019dd923-5e88-73ef-bd5d-0856e1444fb9#slide-99",
  "components": [
    {
      "bbox": null,
      "kind": "callout",
      "text": "GPUs tend to experience failures more frequently than CPUs and all clusters are by no means created equal.",
      "attrs": null,
      "subkind": null,
      "toolName": "Visual emphasis",
      "toolSlug": "visual-emphasis",
      "confidence": null,
      "componentId": "019dd952-47c3-7407-8d17-14642b203d95",
      "frameworkName": null,
      "frameworkSlug": null
    },
    {
      "bbox": {
        "h": 0.5,
        "w": 0.5,
        "x": 0.25,
        "y": 0.45
      },
      "kind": "chart",
      "text": "Root-cause of Interruptions During a 54-day Period of Llama 3 405B Pre-training",
      "attrs": null,
      "subkind": "pie",
      "toolName": null,
      "toolSlug": null,
      "confidence": null,
      "componentId": "7aaf55d2-9f72-4d49-ba58-1512b096d31c",
      "frameworkName": null,
      "frameworkSlug": null
    },
    {
      "bbox": null,
      "kind": "metric",
      "text": "Job interruptions: 8.6",
      "attrs": null,
      "subkind": "primary",
      "toolName": "Quantification",
      "toolSlug": "quantification",
      "confidence": null,
      "componentId": "019dd952-47c3-7407-8d17-19df5ef513a1",
      "frameworkName": null,
      "frameworkSlug": null
    },
    {
      "bbox": {
        "h": 0.2,
        "w": 0.9,
        "x": 0.05,
        "y": 0.22
      },
      "kind": "paragraph",
      "text": "On publishing their Llama 3 family of models, Meta shared a breakdown of the 8.6 job interruptions per day they experienced during a 54-day period of pre-training Llama 3 405B. GPUs tend to experience failures more frequently than CPUs and all clusters are by no means created equal. Continuous monitoring is essential, misconfigurations and dead-on arrival components happen too often due to insufficient testing, and low-cost power, affordable networking rates and availability are paramount. More on power needs in the Politics section!",
      "attrs": null,
      "subkind": "paragraph",
      "toolName": null,
      "toolSlug": null,
      "confidence": null,
      "componentId": "fe7f872d-6c7b-4e8b-bb26-b530ad8ad79c",
      "frameworkName": null,
      "frameworkSlug": null
    },
    {
      "bbox": {
        "h": 0.05,
        "w": 0.85,
        "x": 0.05,
        "y": 0.14
      },
      "kind": "title",
      "text": "But running large clusters continues to be an art and a science of interruptions",
      "attrs": null,
      "subkind": "headline",
      "toolName": null,
      "toolSlug": null,
      "confidence": null,
      "componentId": "a0b762b4-1bb0-4a64-8f65-4f44763e3743",
      "frameworkName": null,
      "frameworkSlug": null
    }
  ],
  "metrics": [],
  "tools": [
    {
      "name": "Action Titles",
      "slug": "action-titles",
      "agent": "Architect",
      "layer": "slide",
      "matchId": "019dd95a-1055-74e0-a959-883dfdee2f06",
      "evidence": "Title 'But running large clusters continues to be an art and a science of interruptions'",
      "confidence": 80
    }
  ],
  "frameworks": [],
  "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": 174,
      "from": 86,
      "beatId": "019dd95a-0682-776c-8e35-73133c63a096",
      "arcName": "Voyage and Return",
      "arcSlug": "voyage-return",
      "beatName": "Discoveries",
      "beatSlug": "voyage-return-discoveries",
      "evidence": "Industry + Politics findings on NVIDIA, regulation, geopolitics",
      "position": 3,
      "confidence": 55,
      "parentBeatName": "Development",
      "parentBeatSlug": "development"
    }
  ],
  "loops": [
    {
      "to": 105,
      "from": 87,
      "name": "David Goliath",
      "slug": "16-david-goliath",
      "bestFor": "Startups, disruptive innovation, competitive displacement",
      "matchId": "019dd95a-07fe-70ce-8d3e-49389fa5a379",
      "evidence": "Pages span $3T market cap, $120B vs $31B counterfactual, compute index, AMD/SoftBank/start-ups failing to dent NVIDIA, smuggling",
      "position": 11,
      "objective": "Establish NVIDIA as unassailable Goliath while challengers fail",
      "structure": "The Giant's Weakness -> Our Slingshot (Unique Edge) -> The Topple",
      "confidence": 82,
      "description": "Frame your initiative as the nimble underdog taking on a slow, bloated incumbent"
    }
  ],
  "imagePathAlt": null,
  "thumbSrc": null,
  "thumbSrcAlt": null,
  "locked": true
}