{
  "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": 20,
  "pageCount": 213,
  "prevPage": 19,
  "nextPage": 21,
  "slideType": "industry_trends",
  "function": "analyze_data",
  "density": "overcrowded",
  "nDataPoints": 20,
  "notes": "The slide critiques the 'vibes-based' evaluation method of the LMSYS leaderboard, noting that GPT-4o and GPT-4o mini share top spots, potentially due to user preference for writing style over raw capability.",
  "elementsJson": [
    "headline_text",
    "bullet_list",
    "bar_chart_vertical"
  ],
  "metadataConfidence": 0.95,
  "imagePath": null,
  "slideHref": "/slides/019dd923-5e88-73ef-bd5d-0856e1444fb9/20",
  "deckHref": "/decks/019dd923-5e88-73ef-bd5d-0856e1444fb9",
  "deckJsonHref": "/decks/019dd923-5e88-73ef-bd5d-0856e1444fb9.json",
  "deckAnchorHref": "/decks/019dd923-5e88-73ef-bd5d-0856e1444fb9#slide-20",
  "components": [
    {
      "bbox": {
        "h": 0.05,
        "w": 0.3,
        "x": 0.65,
        "y": 0.45
      },
      "kind": "callout",
      "text": "Overall: GPT-4o mini joint #1 with GPT-4o!",
      "attrs": null,
      "subkind": "primary",
      "toolName": null,
      "toolSlug": null,
      "confidence": null,
      "componentId": "05f6370a-fc67-481c-8190-6362b0e052f5",
      "frameworkName": null,
      "frameworkSlug": null
    },
    {
      "bbox": {
        "h": 0.5,
        "w": 0.38,
        "x": 0.58,
        "y": 0.4
      },
      "kind": "chart",
      "text": "Figure 1: Confidence Intervals on Model Strength (via Bootstrapping)",
      "attrs": null,
      "subkind": "bar-vertical",
      "toolName": null,
      "toolSlug": null,
      "confidence": null,
      "componentId": "959f4c6f-3015-4c11-93df-a4a4f6e35f78",
      "frameworkName": null,
      "frameworkSlug": null
    },
    {
      "bbox": {
        "h": 0.15,
        "w": 0.55,
        "x": 0.02,
        "y": 0.22
      },
      "kind": "list",
      "text": "The LMSYS Chatbot Arena Leaderboard has emerged as the community's favorite method of formalizing evaluation by 'vibes'. But as model performance improves, it's beginning to produce counterintuitive results.",
      "attrs": null,
      "subkind": "bullet",
      "toolName": null,
      "toolSlug": null,
      "confidence": null,
      "componentId": "5483dbe5-48fb-4caa-a425-4ded69a4d026",
      "frameworkName": null,
      "frameworkSlug": null
    },
    {
      "bbox": {
        "h": 0.1,
        "w": 0.5,
        "x": 0.05,
        "y": 0.67
      },
      "kind": "list",
      "text": "Additionally, as smaller models tend to perform less well on tasks involving more tokens, the 8k context limit arguably gives them an unfair advantage.",
      "attrs": null,
      "subkind": "bullet",
      "toolName": null,
      "toolSlug": null,
      "confidence": null,
      "componentId": "59c0b79e-9ede-4a4f-9b6b-3cfb5bf9b686",
      "frameworkName": null,
      "frameworkSlug": null
    },
    {
      "bbox": {
        "h": 0.1,
        "w": 0.5,
        "x": 0.05,
        "y": 0.43
      },
      "kind": "list",
      "text": "However, controversially, this led to GPT-4o and GPT-4o Mini receiving the same scores, with the latter also outperforming Claude Sonnet 3.5.",
      "attrs": null,
      "subkind": "bullet",
      "toolName": null,
      "toolSlug": null,
      "confidence": null,
      "componentId": "78736fdd-f1b2-4542-a53e-ddb335adc79e",
      "frameworkName": null,
      "frameworkSlug": null
    },
    {
      "bbox": {
        "h": 0.1,
        "w": 0.5,
        "x": 0.05,
        "y": 0.33
      },
      "kind": "list",
      "text": "The arena, which allows users to interact with two randomly selected chatbots side-by-side provides a rough crowdsourced evaluation.",
      "attrs": null,
      "subkind": "bullet",
      "toolName": null,
      "toolSlug": null,
      "confidence": null,
      "componentId": "c6a20903-d1b1-4809-af6e-eecc4f6ef10a",
      "frameworkName": null,
      "frameworkSlug": null
    },
    {
      "bbox": {
        "h": 0.08,
        "w": 0.5,
        "x": 0.05,
        "y": 0.79
      },
      "kind": "list",
      "text": "However, the early version of the vision leaderboard is now beginning to gain traction and aligns better with other evals.",
      "attrs": null,
      "subkind": "bullet",
      "toolName": null,
      "toolSlug": null,
      "confidence": null,
      "componentId": "cad52b0b-085d-4620-ac4f-469da7f0eb04",
      "frameworkName": null,
      "frameworkSlug": null
    },
    {
      "bbox": {
        "h": 0.1,
        "w": 0.5,
        "x": 0.05,
        "y": 0.55
      },
      "kind": "list",
      "text": "This has led to concerns that the ranking is essentially becoming a way of assessing which writing style users happen to prefer most.",
      "attrs": null,
      "subkind": "bullet",
      "toolName": null,
      "toolSlug": null,
      "confidence": null,
      "componentId": "dfa5e230-ca4e-4a1f-9037-f2356a33a979",
      "frameworkName": null,
      "frameworkSlug": null
    },
    {
      "bbox": null,
      "kind": "metric",
      "text": "Model Strength Rating",
      "attrs": null,
      "subkind": "primary",
      "toolName": "Quantification",
      "toolSlug": "quantification",
      "confidence": null,
      "componentId": "019dd952-47c3-7407-8d0f-22091b424a17",
      "frameworkName": null,
      "frameworkSlug": null
    },
    {
      "bbox": {
        "h": 0.05,
        "w": 0.8,
        "x": 0.02,
        "y": 0.13
      },
      "kind": "title",
      "text": "Live by the vibes, die by the vibes...or close your eyes for a year and OpenAI is still #1",
      "attrs": null,
      "subkind": "headline",
      "toolName": null,
      "toolSlug": null,
      "confidence": null,
      "componentId": "a1f710e6-3145-4138-89f0-c554f0456014",
      "frameworkName": null,
      "frameworkSlug": null
    }
  ],
  "metrics": [],
  "tools": [
    {
      "name": "Action Titles",
      "slug": "action-titles",
      "agent": "Architect",
      "layer": "slide",
      "matchId": "019dd95a-1055-74e0-a958-7c2d52d34933",
      "evidence": "Title 'Live by the vibes, die by the vibes...' is opinionated insight",
      "confidence": 80
    },
    {
      "name": "Metaphor & Analogy",
      "slug": "metaphor-analogy",
      "agent": "Storyteller",
      "layer": "slide",
      "matchId": "019dd95a-1055-74e0-a958-820fc53e6342",
      "evidence": "Live/die-by-the-vibes is colloquial metaphor",
      "confidence": 70
    }
  ],
  "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": 20,
      "from": 13,
      "name": "Pattern Hunter",
      "slug": "02-pattern-hunter",
      "bestFor": "Time-pressed audiences, building consensus, when data is strong",
      "matchId": "019dd95a-07fe-70ce-8d3e-27658cc21e8b",
      "evidence": "Series of analyze_data slides on GPT-4, o1, Claude 3.5, Llama 3, openness, contamination, leaderboards",
      "position": 2,
      "objective": "Show frontier model gap is closing through multiple data points",
      "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
}