{
  "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": 140,
  "pageCount": 213,
  "prevPage": 139,
  "nextPage": 141,
  "slideType": "industry_trends",
  "function": "analyze_data",
  "density": "dense",
  "nDataPoints": 4,
  "notes": "The slide discusses the implications of Apple's internal foundation model work on their OpenAI partnership, using Unsloth's performance as a proxy for technical optimization potential.",
  "elementsJson": [
    "headline_text",
    "paragraph",
    "bullet_list",
    "bar_chart_horizontal"
  ],
  "metadataConfidence": 0.95,
  "imagePath": null,
  "slideHref": "/slides/019dd923-5e88-73ef-bd5d-0856e1444fb9/140",
  "deckHref": "/decks/019dd923-5e88-73ef-bd5d-0856e1444fb9",
  "deckJsonHref": "/decks/019dd923-5e88-73ef-bd5d-0856e1444fb9.json",
  "deckAnchorHref": "/decks/019dd923-5e88-73ef-bd5d-0856e1444fb9#slide-140",
  "components": [
    {
      "bbox": null,
      "kind": "callout",
      "text": "Unsloth, since launching at the end of last year, has quickly emerged as a popular open source project, offering radically up to 30x faster training and fine-tuning, by leveraging GPU kernel improvements.",
      "attrs": null,
      "subkind": null,
      "toolName": "Visual emphasis",
      "toolSlug": "visual-emphasis",
      "confidence": null,
      "componentId": "019dd952-47c3-7407-8d1b-f8c7d9ee3952",
      "frameworkName": null,
      "frameworkSlug": null
    },
    {
      "bbox": {
        "h": 0.38,
        "w": 0.42,
        "x": 0.55,
        "y": 0.46
      },
      "kind": "chart",
      "text": "LAION Chip2 210K (1 epoch on 2 T4 GPUs DDP) performance comparison between Unsloth variants and Huggingface.",
      "attrs": null,
      "subkind": "bar-horizontal",
      "toolName": null,
      "toolSlug": null,
      "confidence": null,
      "componentId": "f950cf3f-4319-4f25-9bc4-fd85924e2f50",
      "frameworkName": null,
      "frameworkSlug": null
    },
    {
      "bbox": {
        "h": 0.55,
        "w": 0.53,
        "x": 0.02,
        "y": 0.31
      },
      "kind": "list",
      "text": "Unsloth, since launching at the end of last year, has quickly emerged as a popular open source project, offering radically up to 30x faster training and fine-tuning, by leveraging GPU kernel improvements.\nThe focus is on optimizing the attention mechanism when using LoRA for efficient fine-tuning. Unsloth manually derives gradients for 6 matrix operations, related to LoRA and attention inputs.\nBy carefully arranging the order of matrix multiplications and using in-place operations, it's possible to significantly boost speed and memory efficiency.\nThese optimizations are applied across all model components, not just the attention mechanism.",
      "attrs": null,
      "subkind": "bullet",
      "toolName": null,
      "toolSlug": null,
      "confidence": null,
      "componentId": "b8186f98-4f6b-41e3-a201-c11c81290efc",
      "frameworkName": null,
      "frameworkSlug": null
    },
    {
      "bbox": null,
      "kind": "metric",
      "text": "Training time: 30x",
      "attrs": null,
      "subkind": "primary",
      "toolName": "Quantification",
      "toolSlug": "quantification",
      "confidence": null,
      "componentId": "019dd952-47c3-7407-8d1b-ff55447ab7d5",
      "frameworkName": null,
      "frameworkSlug": null
    },
    {
      "bbox": {
        "h": 0.08,
        "w": 0.96,
        "x": 0.02,
        "y": 0.21
      },
      "kind": "paragraph",
      "text": "Given Apple is publishing work on foundation models that will power Apple Intelligence features, it's reasonable to ask how long-lasting or deep any OpenAI partnership is likely to be.",
      "attrs": null,
      "subkind": "paragraph",
      "toolName": null,
      "toolSlug": null,
      "confidence": null,
      "componentId": "adff971e-f961-414f-8f9b-330b0c3a4ffe",
      "frameworkName": null,
      "frameworkSlug": null
    },
    {
      "bbox": {
        "h": 0.04,
        "w": 0.32,
        "x": 0.02,
        "y": 0.14
      },
      "kind": "title",
      "text": "There's gold in them kernels",
      "attrs": null,
      "subkind": "headline",
      "toolName": null,
      "toolSlug": null,
      "confidence": null,
      "componentId": "ba290d8d-3c6b-4b1b-bf3d-4345b95da574",
      "frameworkName": null,
      "frameworkSlug": null
    }
  ],
  "metrics": [],
  "tools": [
    {
      "name": "Chart Selection Guide",
      "slug": "chart-selection-guide",
      "agent": "Designer",
      "layer": "slide",
      "matchId": "23b9a32d-c38b-4f6d-b692-8bdd9ddd5893",
      "evidence": "The slide contains a chart/bar-horizontal component showing performance comparison between Unsloth variants and Huggingface.",
      "confidence": 0.6
    }
  ],
  "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": 147,
      "from": 137,
      "name": "Pattern Hunter",
      "slug": "02-pattern-hunter",
      "bestFor": "Time-pressed audiences, building consensus, when data is strong",
      "matchId": "019dd95a-07fe-70ce-8d3e-5d103aecde20",
      "evidence": "Law 90%, Apple+OpenAI, Unsloth, Recursion-Exscientia, video gen race, Kling, mRNA, Ray-Ban, Rabbit R1",
      "position": 16,
      "objective": "Show AI penetration into law/medicine/devices/glasses verticals",
      "structure": "Evidence A -> Evidence B -> Evidence C -> Pattern/Conclusion",
      "confidence": 65,
      "description": "Group multiple pieces of evidence that together point to a pattern or conclusion"
    }
  ],
  "imagePathAlt": null,
  "thumbSrc": null,
  "thumbSrcAlt": null,
  "locked": true
}