{
  "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": 146,
  "pageCount": 163,
  "prevPage": 145,
  "nextPage": 147,
  "slideType": "industry_trends",
  "function": "summarize",
  "density": "dense",
  "nDataPoints": 0,
  "notes": "Part of the State of AI 2023 report.",
  "elementsJson": [
    "headline_text",
    "bullet_list",
    "paragraph"
  ],
  "metadataConfidence": 1,
  "imagePath": null,
  "slideHref": "/slides/019dd923-5e88-73ef-bd5d-06b04d219fea/146",
  "deckHref": "/decks/019dd923-5e88-73ef-bd5d-06b04d219fea",
  "deckJsonHref": "/decks/019dd923-5e88-73ef-bd5d-06b04d219fea.json",
  "deckAnchorHref": "/decks/019dd923-5e88-73ef-bd5d-06b04d219fea#slide-146",
  "components": [
    {
      "bbox": null,
      "kind": "callout",
      "text": "Open source LLMs level the playing field for research and enterprises but come with higher risk of proliferation and misuse by bad actors. Closed source APIs offers more security and control but less transparency.",
      "attrs": null,
      "subkind": null,
      "toolName": "Visual emphasis",
      "toolSlug": "visual-emphasis",
      "confidence": null,
      "componentId": "019dd952-47c4-7719-a22f-99d8a2078a17",
      "frameworkName": null,
      "frameworkSlug": null
    },
    {
      "bbox": {
        "h": 0.18,
        "w": 0.96,
        "x": 0.02,
        "y": 0.28
      },
      "kind": "list",
      "text": "The approach to open source safety differs among companies with no standard guidelines. Meta's release of Llama2 came with an extensive overview of safety measures and a Responsible Use Guide to provide best practices for developers. In contrast, Adept's release of the Persimmon 8B model skipped safety entirely: \"we have not added further finetuning, postprocessing or sampling strategies to control for toxic outputs.\"",
      "attrs": null,
      "subkind": "bullet",
      "toolName": null,
      "toolSlug": null,
      "confidence": null,
      "componentId": "3bd3a136-bab9-44c1-bf2d-4825a890a58a",
      "frameworkName": null,
      "frameworkSlug": null
    },
    {
      "bbox": {
        "h": 0.18,
        "w": 0.96,
        "x": 0.02,
        "y": 0.73
      },
      "kind": "list",
      "text": "API-based LLM misuse is easier to curtail through iterative deployment. OpenAI has internal detection and response infrastructure to handle misuse of the APIs based on their usage policy as well as responding to real world scenarios (e.g. spam promotions for dubious medical products). With GPT3.5 turbo fine-tuning capability, training data is filtered using OpenAI's moderation API to preserve default model safety.",
      "attrs": null,
      "subkind": "bullet",
      "toolName": null,
      "toolSlug": null,
      "confidence": null,
      "componentId": "664dc5d5-c43a-41fa-9ad9-07610c73dd66",
      "frameworkName": null,
      "frameworkSlug": null
    },
    {
      "bbox": {
        "h": 0.25,
        "w": 0.96,
        "x": 0.02,
        "y": 0.47
      },
      "kind": "list",
      "text": "To download Llama2 weights, users need to sign an agreement stating their intent not to use it for malicious purposes, however it's unclear who will enforce this. Models distributed via Hugging Face have licenses that restrict usage and offer moderation. Fine tuning models for malicious use opens a pandora's box of misuse e.g. \"WormGPT\" to aid cybercrime (albeit using an older GPT-J model with poor performance). We've yet seen scaled proliferation of small models (~8B size) fine tuned for misuse and optimized for on-device inference.",
      "attrs": null,
      "subkind": "bullet",
      "toolName": null,
      "toolSlug": null,
      "confidence": null,
      "componentId": "ee89ec32-e200-4963-b3d2-82baaeaa777d",
      "frameworkName": null,
      "frameworkSlug": null
    },
    {
      "bbox": {
        "h": 0.08,
        "w": 0.96,
        "x": 0.02,
        "y": 0.19
      },
      "kind": "paragraph",
      "text": "Open source LLMs level the playing field for research and enterprises but come with higher risk of proliferation and misuse by bad actors. Closed source APIs offers more security and control but less transparency.",
      "attrs": null,
      "subkind": "paragraph",
      "toolName": null,
      "toolSlug": null,
      "confidence": null,
      "componentId": "2fa277e0-716d-4501-8b55-5a3ec6ff9c54",
      "frameworkName": null,
      "frameworkSlug": null
    },
    {
      "bbox": null,
      "kind": "quote",
      "text": "we have not added further finetuning, postprocessing or sampling strategies to control for toxic outputs.",
      "attrs": null,
      "subkind": null,
      "toolName": "Authority citation",
      "toolSlug": "authority-citation",
      "confidence": null,
      "componentId": "019dd952-47c4-7719-a22f-9deedb33639f",
      "frameworkName": null,
      "frameworkSlug": null
    },
    {
      "bbox": {
        "h": 0.03,
        "w": 0.45,
        "x": 0.02,
        "y": 0.14
      },
      "kind": "title",
      "text": "The open vs. closed source debate continues..",
      "attrs": null,
      "subkind": "headline",
      "toolName": null,
      "toolSlug": null,
      "confidence": null,
      "componentId": "f2112250-6969-45e8-8a15-04e60007b094",
      "frameworkName": null,
      "frameworkSlug": null
    }
  ],
  "metrics": [],
  "tools": [
    {
      "name": "Contrast Pairs",
      "slug": "contrast-pairs",
      "agent": "Storyteller",
      "layer": "loop",
      "matchId": "019dd95a-1055-74e0-a958-0446b7af1486",
      "evidence": "Open vs. closed source paired explicitly with risk vs. control trade-off.",
      "confidence": 75
    },
    {
      "name": "Action Titles",
      "slug": "action-titles",
      "agent": "Architect",
      "layer": "slide",
      "matchId": "019dd95a-1055-74e0-a958-02e136821761",
      "evidence": "Title 'The open vs. closed source debate continues' frames the trade-off.",
      "confidence": 75
    }
  ],
  "frameworks": [],
  "arcBeats": [
    {
      "to": 155,
      "from": 121,
      "beatId": "019dd95a-0682-776c-8e35-478c8c58938c",
      "arcName": "The Triple Take",
      "arcSlug": "triple-take",
      "beatName": "The Implications (So What)",
      "beatSlug": "triple-take-the-implications-so-what",
      "evidence": "Politics + Safety sections explore regulatory, geopolitical and risk implications.",
      "position": 2,
      "confidence": 78,
      "parentBeatName": "Reflection",
      "parentBeatSlug": "reflection"
    },
    {
      "to": 155,
      "from": 121,
      "beatId": "019dd95a-0682-776c-8e35-554ec1797b3a",
      "arcName": "The Mountain",
      "arcSlug": "mountain",
      "beatName": "Climax",
      "beatSlug": "mountain-climax",
      "evidence": "Regulatory divergence and the x-risk debate erupt into the mainstream.",
      "position": 3,
      "confidence": 45,
      "parentBeatName": "Turn",
      "parentBeatSlug": "turn"
    }
  ],
  "loops": [
    {
      "to": 148,
      "from": 139,
      "name": "Pattern Hunter",
      "slug": "02-pattern-hunter",
      "bestFor": "Time-pressed audiences, building consensus, when data is strong",
      "matchId": "019dd95a-07fe-70ce-8d3e-17c0a044d464",
      "evidence": "Hendrycks taxonomy, x-risk explosion, sceptics, govt attention, jailbreak evidence accumulate.",
      "position": 14,
      "objective": "Frame the safety debate, both x-risk and demonstrated unsafe behaviour",
      "structure": "Evidence A -> Evidence B -> Evidence C -> Pattern/Conclusion",
      "confidence": 78,
      "description": "Group multiple pieces of evidence that together point to a pattern or conclusion"
    }
  ],
  "imagePathAlt": null,
  "thumbSrc": null,
  "thumbSrcAlt": null,
  "locked": true
}