{
  "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": 25,
  "pageCount": 213,
  "prevPage": 24,
  "nextPage": 26,
  "slideType": "industry_trends",
  "function": "summarize",
  "density": "dense",
  "nDataPoints": 3,
  "notes": "Includes a visual comparison of image compression levels (32 tokens vs 256 tokens vs 65536 pixels).",
  "elementsJson": [
    "headline_text",
    "bullet_list",
    "infographic",
    "paragraph"
  ],
  "metadataConfidence": 1,
  "imagePath": null,
  "slideHref": "/slides/019dd923-5e88-73ef-bd5d-0856e1444fb9/25",
  "deckHref": "/decks/019dd923-5e88-73ef-bd5d-0856e1444fb9",
  "deckJsonHref": "/decks/019dd923-5e88-73ef-bd5d-0856e1444fb9.json",
  "deckAnchorHref": "/decks/019dd923-5e88-73ef-bd5d-0856e1444fb9#slide-25",
  "components": [
    {
      "bbox": null,
      "kind": "callout",
      "text": "It's possible to shrink the memory requirements of LLMs by reducing the precision of their parameters.",
      "attrs": null,
      "subkind": null,
      "toolName": "Visual emphasis",
      "toolSlug": "visual-emphasis",
      "confidence": null,
      "componentId": "019dd952-47c3-7407-8d0f-a70d6eece685",
      "frameworkName": null,
      "frameworkSlug": null
    },
    {
      "bbox": {
        "h": 0.6,
        "w": 0.25,
        "x": 0.7,
        "y": 0.3
      },
      "kind": "image",
      "text": "Latent size and costs comparison showing bear images at 32 tokens, 256 tokens, and 65536 pixels.",
      "attrs": null,
      "subkind": "infographic",
      "toolName": null,
      "toolSlug": null,
      "confidence": null,
      "componentId": "e117401e-707a-4811-9c7d-ba8c700a8ffe",
      "frameworkName": null,
      "frameworkSlug": null
    },
    {
      "bbox": {
        "h": 0.55,
        "w": 0.65,
        "x": 0.03,
        "y": 0.3
      },
      "kind": "list",
      "text": "Microsoft's BitNet uses a \"BitLinear\" layer to replace standard linear layers, employing 1-bit weights and quantized activations.\nIt shows competitive performance compared to full-precision models and demonstrates a scaling law similar to full-precision transformers, with significant memory and energy savings.\nMicrosoft followed up with BitNet b1.58, with ternary weights to match full-precision LLM performance at 3B size while retaining efficiency gains.\nMeanwhile, ByteDance's TiTok (Transformer-based 1-Dimensional Tokenizer) quantizes images into compact 1D sequences of discrete token for image reconstruction and generation tasks. This allows images to be represented with as few as 32 tokens, instead of hundreds or thousands.",
      "attrs": null,
      "subkind": "bullet",
      "toolName": null,
      "toolSlug": null,
      "confidence": null,
      "componentId": "5de0ad5c-3679-40e8-bec1-ecb463fb787c",
      "frameworkName": null,
      "frameworkSlug": null
    },
    {
      "bbox": {
        "h": 0.08,
        "w": 0.85,
        "x": 0.03,
        "y": 0.21
      },
      "kind": "paragraph",
      "text": "It's possible to shrink the memory requirements of LLMs by reducing the precision of their parameters. Researchers are increasingly managing to minimize the performance trade-offs.",
      "attrs": null,
      "subkind": "paragraph",
      "toolName": null,
      "toolSlug": null,
      "confidence": null,
      "componentId": "20b4591a-e603-4ade-a80c-f0a95d48c9dc",
      "frameworkName": null,
      "frameworkSlug": null
    },
    {
      "bbox": {
        "h": 0.05,
        "w": 0.6,
        "x": 0.03,
        "y": 0.13
      },
      "kind": "title",
      "text": "Strong results in quantization point to an on-device future",
      "attrs": null,
      "subkind": "headline",
      "toolName": null,
      "toolSlug": null,
      "confidence": null,
      "componentId": "b90e45a9-e092-4576-8c7d-b3ba2bb6e76d",
      "frameworkName": null,
      "frameworkSlug": null
    }
  ],
  "metrics": [],
  "tools": [
    {
      "name": "Action Titles",
      "slug": "action-titles",
      "agent": "Architect",
      "layer": "slide",
      "matchId": "019dd95a-1055-74e0-a958-9915d90ab6f0",
      "evidence": "Title 'Strong results in quantization point to an on-device future'",
      "confidence": 88
    },
    {
      "name": "Data-Ink Ratio",
      "slug": "data-ink-ratio",
      "agent": "Designer",
      "layer": "slide",
      "matchId": "be452ee9-f22b-454d-8c41-564a9f2964e1",
      "evidence": "Latent size and costs comparison showing bear images at 32 tokens, 256 tokens, and 65536 pixels.",
      "confidence": 0.7
    }
  ],
  "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": 85,
      "from": 12,
      "beatId": "019dd95a-0682-776c-8e35-6d9d8ee06d0a",
      "arcName": "Voyage and Return",
      "arcSlug": "voyage-return",
      "beatName": "The Unknown",
      "beatSlug": "voyage-return-the-unknown",
      "evidence": "Research section explores frontier model uncharted territory",
      "position": 2,
      "confidence": 55,
      "parentBeatName": "Complication",
      "parentBeatSlug": "complication"
    }
  ],
  "loops": [
    {
      "to": 29,
      "from": 21,
      "name": "Pattern Hunter",
      "slug": "02-pattern-hunter",
      "bestFor": "Time-pressed audiences, building consensus, when data is strong",
      "matchId": "019dd95a-07fe-70ce-8d3e-2b2229583eee",
      "evidence": "Eight slides on shrinking models, on-device, hybrids, transformers reigning",
      "position": 3,
      "objective": "Demonstrate efficiency frontier across distillation, quantization, hybrids",
      "structure": "Evidence A -> Evidence B -> Evidence C -> Pattern/Conclusion",
      "confidence": 75,
      "description": "Group multiple pieces of evidence that together point to a pattern or conclusion"
    }
  ],
  "imagePathAlt": null,
  "thumbSrc": null,
  "thumbSrcAlt": null,
  "locked": true
}