{
  "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": 65,
  "pageCount": 163,
  "prevPage": 64,
  "nextPage": 66,
  "slideType": "case_study",
  "function": "illustrate_case",
  "density": "dense",
  "nDataPoints": 1,
  "notes": "The slide highlights the use of social media data for specialized medical AI training.",
  "elementsJson": [
    "headline_text",
    "paragraph",
    "bullet_list",
    "process_diagram",
    "data_table"
  ],
  "metadataConfidence": 0.95,
  "imagePath": null,
  "slideHref": "/slides/019dd923-5e88-73ef-bd5d-06b04d219fea/65",
  "deckHref": "/decks/019dd923-5e88-73ef-bd5d-06b04d219fea",
  "deckJsonHref": "/decks/019dd923-5e88-73ef-bd5d-06b04d219fea.json",
  "deckAnchorHref": "/decks/019dd923-5e88-73ef-bd5d-06b04d219fea#slide-65",
  "components": [
    {
      "bbox": null,
      "kind": "callout",
      "text": "Compared to CLIP, PLIP has 2-6x better Precision@10.",
      "attrs": null,
      "subkind": null,
      "toolName": "Visual emphasis",
      "toolSlug": "visual-emphasis",
      "confidence": null,
      "componentId": "019dd952-47c4-7719-a226-5b9c2eff7a8a",
      "frameworkName": null,
      "frameworkSlug": null
    },
    {
      "bbox": {
        "h": 0.22,
        "w": 0.22,
        "x": 0.75,
        "y": 0.68
      },
      "kind": "chart",
      "text": "Precision@10 comparison chart",
      "attrs": null,
      "subkind": "bar-horizontal",
      "toolName": null,
      "toolSlug": null,
      "confidence": null,
      "componentId": "83362e19-e5a4-42b4-adff-1064e876f7f9",
      "frameworkName": null,
      "frameworkSlug": null
    },
    {
      "bbox": {
        "h": 0.48,
        "w": 0.42,
        "x": 0.55,
        "y": 0.42
      },
      "kind": "image",
      "text": "Diagram showing the data collection process from Twitter and the model architecture.",
      "attrs": null,
      "subkind": "infographic",
      "toolName": null,
      "toolSlug": null,
      "confidence": null,
      "componentId": "757e5086-1589-4f27-b254-4ba2b1fed8d0",
      "frameworkName": null,
      "frameworkSlug": null
    },
    {
      "bbox": {
        "h": 0.45,
        "w": 0.5,
        "x": 0.02,
        "y": 0.41
      },
      "kind": "list",
      "text": "Like CLIP, PLIP can perform zero-shot classification on unseen data, enabling it to distin several key tissue types.\nIt can also be used to improve text-to-image and image-to-image retrieval of pathology images.\nUnlike other machine learning approaches in digital pathology that are predicated on learning from a fixed set of labels, PLIP can be more generally applied and is flexible to the changing nature of diagnostic criteria in pathology.\nCompared to CLIP, PLIP has 2-6x better Precision@10.",
      "attrs": null,
      "subkind": "bullet",
      "toolName": null,
      "toolSlug": null,
      "confidence": null,
      "componentId": "a3a99950-7916-4dfa-bf9b-be30abd9f510",
      "frameworkName": null,
      "frameworkSlug": null
    },
    {
      "bbox": null,
      "kind": "metric",
      "text": "Precision@10: 2-6x",
      "attrs": null,
      "subkind": "primary",
      "toolName": "Quantification",
      "toolSlug": "quantification",
      "confidence": null,
      "componentId": "019dd952-47c4-7719-a226-5e05b34ce656",
      "frameworkName": null,
      "frameworkSlug": null
    },
    {
      "bbox": {
        "h": 0.15,
        "w": 0.95,
        "x": 0.02,
        "y": 0.21
      },
      "kind": "paragraph",
      "text": "It's no secret that (quality) data is king for building capable AI systems, and no more so than in domains such as clinical medicine where (quality) data is expensive to produce. This work mines text-image pairs on Twitter to create the OpenPath dataset with 200+ pathology images paired with natural language descriptors. Inspired by OpenAI's Contrastive Language-Image Pretraining (CLIP) model, the authors create P(athology)LIP.",
      "attrs": null,
      "subkind": "paragraph",
      "toolName": null,
      "toolSlug": null,
      "confidence": null,
      "componentId": "aecd34d8-f539-408d-a410-7a273b33a548",
      "frameworkName": null,
      "frameworkSlug": null
    },
    {
      "bbox": {
        "h": 0.05,
        "w": 0.95,
        "x": 0.02,
        "y": 0.13
      },
      "kind": "title",
      "text": "Tweet storm: a SOTA pathology language-image pretrained model from medical Twitter",
      "attrs": null,
      "subkind": "headline",
      "toolName": null,
      "toolSlug": null,
      "confidence": null,
      "componentId": "c80219f3-2c60-4541-9a08-54529b40ad3a",
      "frameworkName": null,
      "frameworkSlug": null
    }
  ],
  "metrics": [],
  "tools": [
    {
      "name": "Metaphor & Analogy",
      "slug": "metaphor-analogy",
      "agent": "Storyteller",
      "layer": "slide",
      "matchId": "ac22f2fe-39d7-4502-8263-84fe412f595c",
      "evidence": "Like CLIP, PLIP can perform zero-shot classification on unseen data, enabling it to distin several key tissue types.",
      "confidence": 0.6
    }
  ],
  "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": 68,
      "from": 59,
      "name": "Pattern Hunter",
      "slug": "02-pattern-hunter",
      "bestFor": "Time-pressed audiences, building consensus, when data is strong",
      "matchId": "019dd95a-07fe-70ce-8d3d-f9a066ea1e78",
      "evidence": "RFdiffusion, ESMFold, PerturbNet, AlphaMissense, Med-PaLM, PLIP, CoDoC, then meta map of AI-for-science fields.",
      "position": 7,
      "objective": "AI for science accumulates breakthroughs across bio and medicine",
      "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
}