{
  "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": 30,
  "pageCount": 213,
  "prevPage": 29,
  "nextPage": 31,
  "slideType": "industry_trends",
  "function": "summarize",
  "density": "overcrowded",
  "nDataPoints": 2,
  "notes": "The slide includes a process diagram for Nemotron-4-340B and a feature comparison table for various AI infrastructure providers.",
  "elementsJson": [
    "headline_text",
    "bullet_list",
    "flowchart",
    "comparison_table"
  ],
  "metadataConfidence": 0.95,
  "imagePath": null,
  "slideHref": "/slides/019dd923-5e88-73ef-bd5d-0856e1444fb9/30",
  "deckHref": "/decks/019dd923-5e88-73ef-bd5d-0856e1444fb9",
  "deckJsonHref": "/decks/019dd923-5e88-73ef-bd5d-0856e1444fb9.json",
  "deckAnchorHref": "/decks/019dd923-5e88-73ef-bd5d-0856e1444fb9#slide-30",
  "components": [
    {
      "bbox": null,
      "kind": "callout",
      "text": "Opinion seems to be warming.",
      "attrs": null,
      "subkind": null,
      "toolName": "Visual emphasis",
      "toolSlug": "visual-emphasis",
      "confidence": null,
      "componentId": "019dd952-47c3-7407-8d11-d46815e3bdbc",
      "frameworkName": null,
      "frameworkSlug": null
    },
    {
      "bbox": {
        "h": 0.2,
        "w": 0.3,
        "x": 0.68,
        "y": 0.43
      },
      "kind": "diagram",
      "text": "Nemotron-4-340B synthetic data generation workflow",
      "attrs": null,
      "subkind": "process",
      "toolName": null,
      "toolSlug": null,
      "confidence": null,
      "componentId": "e6b10941-3ef0-4c72-aae2-1511e4779d32",
      "frameworkName": null,
      "frameworkSlug": null
    },
    {
      "bbox": {
        "h": 0.1,
        "w": 0.65,
        "x": 0.02,
        "y": 0.53
      },
      "kind": "list",
      "text": "To make this process easier, NVIDIA released the Nemotron-4-340B family, a suite of models designed specifically for synthetic data generation, available via a permissive license. Meta's Llama can also be used for synthetic data generation.",
      "attrs": null,
      "subkind": "bullet",
      "toolName": null,
      "toolSlug": null,
      "confidence": null,
      "componentId": "103520a5-58de-4b8d-ad10-ac742317948b",
      "frameworkName": null,
      "frameworkSlug": null
    },
    {
      "bbox": {
        "h": 0.08,
        "w": 0.65,
        "x": 0.02,
        "y": 0.31
      },
      "kind": "list",
      "text": "As well as being the main source of training data for the Phi family, synthetic data was used by Anthropic when training Claude 3 to help represent scenarios that might have been missing in the training data.",
      "attrs": null,
      "subkind": "bullet",
      "toolName": null,
      "toolSlug": null,
      "confidence": null,
      "componentId": "1e04ad86-e0e6-4c6d-9ad0-ca88a1e9826f",
      "frameworkName": null,
      "frameworkSlug": null
    },
    {
      "bbox": {
        "h": 0.08,
        "w": 0.65,
        "x": 0.02,
        "y": 0.65
      },
      "kind": "list",
      "text": "It also appears possible to create synthetic high-quality instruction data by extracting it directly from an aligned LLM, with techniques like Magpie. Models fine-tuned this way sometimes perform comparably to Llama-3-8B-Instruct.",
      "attrs": null,
      "subkind": "bullet",
      "toolName": null,
      "toolSlug": null,
      "confidence": null,
      "componentId": "8c21bf31-19a0-43fb-9e3f-1e8bad9a5bb6",
      "frameworkName": null,
      "frameworkSlug": null
    },
    {
      "bbox": {
        "h": 0.1,
        "w": 0.65,
        "x": 0.02,
        "y": 0.41
      },
      "kind": "list",
      "text": "Hugging Face used Mixtral-8x7B Instruct to generate over 30M files and 25B tokens of synthetic textbooks, blog posts, and stories to recreate the Phi-1.5 training dataset, which they dubbed Cosmopedia.",
      "attrs": null,
      "subkind": "bullet",
      "toolName": null,
      "toolSlug": null,
      "confidence": null,
      "componentId": "9252c260-e4e4-4fa1-93c5-8e112a7c8055",
      "frameworkName": null,
      "frameworkSlug": null
    },
    {
      "bbox": {
        "h": 0.08,
        "w": 0.65,
        "x": 0.02,
        "y": 0.22
      },
      "kind": "list",
      "text": "Last year's report pointed to the divides of opinion around synthetic data: with some finding it useful, others fearing its potential to trigger model collapse by compounding errors. Opinion seems to be warming.",
      "attrs": null,
      "subkind": "bullet",
      "toolName": null,
      "toolSlug": null,
      "confidence": null,
      "componentId": "9eecbdf0-d464-47bc-8ac8-cccb986e8fe8",
      "frameworkName": null,
      "frameworkSlug": null
    },
    {
      "bbox": {
        "h": 0.25,
        "w": 0.3,
        "x": 0.68,
        "y": 0.68
      },
      "kind": "table",
      "text": "Feature comparison table for AI infrastructure providers",
      "attrs": null,
      "subkind": "data",
      "toolName": null,
      "toolSlug": null,
      "confidence": null,
      "componentId": "9c67ad2b-05af-47ca-b4ec-beef26d3985a",
      "frameworkName": null,
      "frameworkSlug": null
    },
    {
      "bbox": {
        "h": 0.05,
        "w": 0.55,
        "x": 0.02,
        "y": 0.14
      },
      "kind": "title",
      "text": "Synthetic data starts gaining more widespread adoption...",
      "attrs": null,
      "subkind": "headline",
      "toolName": null,
      "toolSlug": null,
      "confidence": null,
      "componentId": "f8387404-b721-48e5-812d-f88b316291de",
      "frameworkName": null,
      "frameworkSlug": null
    }
  ],
  "metrics": [],
  "tools": [
    {
      "name": "Action Titles",
      "slug": "action-titles",
      "agent": "Architect",
      "layer": "slide",
      "matchId": "019dd95a-1055-74e0-a958-a6f555c6159a",
      "evidence": "Title 'Synthetic data starts gaining more widespread adoption...'",
      "confidence": 80
    },
    {
      "name": "List presentation",
      "slug": "list-presentation",
      "agent": null,
      "layer": "slide",
      "matchId": "83fa60a8-6225-4c4e-82b5-aa67d8b35f32",
      "evidence": "Feature comparison table for AI infrastructure providers",
      "confidence": 0.5
    }
  ],
  "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": 37,
      "from": 30,
      "name": "Pattern Hunter",
      "slug": "02-pattern-hunter",
      "bestFor": "Time-pressed audiences, building consensus, when data is strong",
      "matchId": "019dd95a-07fe-70ce-8d3e-2e9bfb3e3bea",
      "evidence": "Synthetic data, FineWeb, embeddings, RAG, power grid, JEST data curation",
      "position": 4,
      "objective": "Survey data, retrieval and training-infra advances",
      "structure": "Evidence A -> Evidence B -> Evidence C -> Pattern/Conclusion",
      "confidence": 72,
      "description": "Group multiple pieces of evidence that together point to a pattern or conclusion"
    }
  ],
  "imagePathAlt": null,
  "thumbSrc": null,
  "thumbSrcAlt": null,
  "locked": true
}