{
  "docId": "019dd923-5e88-73ef-bd5d-06b04d219fea",
  "docSlug": "dd91c78f6570bf29",
  "documentTitle": "2023 Air Street Capital The State of AI Report 2023",
  "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": 18,
  "pageCount": 163,
  "prevPage": 17,
  "nextPage": 19,
  "slideType": "industry_trends",
  "function": "summarize",
  "density": "dense",
  "nDataPoints": 0,
  "notes": "The slide discusses the ecosystem of open-source LLMs, mentioning specific models like MPT-30B, Falcon, RedPajama, Pythia, and Vicuna, and techniques like LoRA.",
  "elementsJson": [
    "headline_text",
    "bullet_list",
    "paragraph",
    "photo"
  ],
  "metadataConfidence": 0.95,
  "imagePath": null,
  "slideHref": "/slides/019dd923-5e88-73ef-bd5d-06b04d219fea/18",
  "deckHref": "/decks/019dd923-5e88-73ef-bd5d-06b04d219fea",
  "deckJsonHref": "/decks/019dd923-5e88-73ef-bd5d-06b04d219fea.json",
  "deckAnchorHref": "/decks/019dd923-5e88-73ef-bd5d-06b04d219fea#slide-18",
  "components": [
    {
      "bbox": null,
      "kind": "callout",
      "text": "After Meta released LLaMa-1, other institutions joined the movement to release the weights of relatively large language models.",
      "attrs": null,
      "subkind": null,
      "toolName": "Visual emphasis",
      "toolSlug": "visual-emphasis",
      "confidence": null,
      "componentId": "019dd952-47c4-7719-a221-f5c1e58f3063",
      "frameworkName": null,
      "frameworkSlug": null
    },
    {
      "bbox": {
        "h": 0.35,
        "w": 0.25,
        "x": 0.7,
        "y": 0.48
      },
      "kind": "image",
      "text": null,
      "attrs": null,
      "subkind": "illustration",
      "toolName": null,
      "toolSlug": null,
      "confidence": null,
      "componentId": "82a9ac6f-fa5a-4fe2-a62c-4da97c3c0a81",
      "frameworkName": null,
      "frameworkSlug": null
    },
    {
      "bbox": {
        "h": 0.18,
        "w": 0.65,
        "x": 0.02,
        "y": 0.67
      },
      "kind": "list",
      "text": "Helped with parameter-efficient fine-tuning methods like LoRA (Low-rank adaptation of LLMs – initially by Microsoft), LM practitioners started fine-tuning these pre-trained LLMs for specific applications like (of course) chat. One example is LMSys’s Vicuna which is LLaMa fine-tuned on user-shared conversations with ChatGPT.",
      "attrs": null,
      "subkind": "bullet",
      "toolName": null,
      "toolSlug": null,
      "confidence": null,
      "componentId": "936cd334-ec4f-4db4-b53e-5e20ce25cf5a",
      "frameworkName": null,
      "frameworkSlug": null
    },
    {
      "bbox": {
        "h": 0.18,
        "w": 0.65,
        "x": 0.02,
        "y": 0.48
      },
      "kind": "list",
      "text": "Notably, RedPajama aimed to exactly replicate LLaMa-1 to make it fully open-source. Falcon 40B came from a new entrant in the LLM sweepstakes, TII UAE, and was quickly made open-source. Falcon-180B was later released, but was notably trained on very little code, and not tested on coding.",
      "attrs": null,
      "subkind": "bullet",
      "toolName": null,
      "toolSlug": null,
      "confidence": null,
      "componentId": "db97b753-587b-4646-a0e0-8b405e497630",
      "frameworkName": null,
      "frameworkSlug": null
    },
    {
      "bbox": {
        "h": 0.25,
        "w": 0.95,
        "x": 0.02,
        "y": 0.22
      },
      "kind": "paragraph",
      "text": "After Meta released LLaMa-1, other institutions joined the movement to release the weights of relatively large language models. A few of them stand out, like MosaicML’s MPT-30B, TII UAE’s Falcon-40B, Together’s RedPajama, or Eleuther’s Pythia. Meanwhile another dynamic was taking place, where the open-source community fine-tuned the smallest versions of LLaMa on specialized datasets and applied them to dozens of downstream applications. Mistral AI’s 7B model also recently emerged as the strongest small model.",
      "attrs": null,
      "subkind": "paragraph",
      "toolName": null,
      "toolSlug": null,
      "confidence": null,
      "componentId": "5c04b3a5-0c61-4321-b445-419f07c366c4",
      "frameworkName": null,
      "frameworkSlug": null
    },
    {
      "bbox": {
        "h": 0.05,
        "w": 0.7,
        "x": 0.02,
        "y": 0.14
      },
      "kind": "title",
      "text": "LLaMa sets off a race of open(ish) competitive Large Language Models",
      "attrs": null,
      "subkind": "headline",
      "toolName": null,
      "toolSlug": null,
      "confidence": null,
      "componentId": "3dcf5b66-fb97-4825-8699-eac770ad8f9a",
      "frameworkName": null,
      "frameworkSlug": null
    }
  ],
  "metrics": [],
  "tools": [],
  "frameworks": [],
  "arcBeats": [
    {
      "to": 120,
      "from": 11,
      "beatId": "019dd95a-0682-776c-8e35-41afd44ef59f",
      "arcName": "The Triple Take",
      "arcSlug": "triple-take",
      "beatName": "The Facts (What)",
      "beatSlug": "triple-take-the-facts-what",
      "evidence": "Research + Industry sections inventory model, compute, funding facts.",
      "position": 1,
      "confidence": 78,
      "parentBeatName": "Setup",
      "parentBeatSlug": "setup"
    },
    {
      "to": 120,
      "from": 11,
      "beatId": "019dd95a-0682-776c-8e35-523bfb7f96e6",
      "arcName": "The Mountain",
      "arcSlug": "mountain",
      "beatName": "Rising Action",
      "beatSlug": "mountain-rising-action",
      "evidence": "Escalating capabilities, compute concentration and capital flows.",
      "position": 2,
      "confidence": 45,
      "parentBeatName": "Development",
      "parentBeatSlug": "development"
    }
  ],
  "loops": [
    {
      "to": 21,
      "from": 12,
      "name": "Pattern Hunter",
      "slug": "02-pattern-hunter",
      "bestFor": "Time-pressed audiences, building consensus, when data is strong",
      "matchId": "019dd95a-07fe-70ce-8d3d-e61d00ffc6e4",
      "evidence": "Sequential GPT-4, RLHF, LLaMa, popularity-mention slides converge on a single conclusion.",
      "position": 2,
      "objective": "Pile up evidence that GPT-4 and RLHF dominate the LLM landscape",
      "structure": "Evidence A -> Evidence B -> Evidence C -> Pattern/Conclusion",
      "confidence": 82,
      "description": "Group multiple pieces of evidence that together point to a pattern or conclusion"
    }
  ],
  "imagePathAlt": null,
  "thumbSrc": null,
  "thumbSrcAlt": null,
  "locked": true
}