{
  "docId": "019dd923-5e88-73ef-bd5d-0d0f98caffe1",
  "docSlug": "5df6bd1b0447b5f6",
  "documentTitle": "2025 Air Street Capital The State of AI Report 2025",
  "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": 17,
  "pageCount": 313,
  "prevPage": 16,
  "nextPage": 18,
  "slideType": "industry_trends",
  "function": "present_solution",
  "density": "overcrowded",
  "nDataPoints": 2,
  "notes": "Discusses technical advancements in LLM inference strategies.",
  "elementsJson": [
    "headline_text",
    "bullet_list",
    "process_diagram"
  ],
  "metadataConfidence": 1,
  "imagePath": null,
  "slideHref": "/slides/019dd923-5e88-73ef-bd5d-0d0f98caffe1/17",
  "deckHref": "/decks/019dd923-5e88-73ef-bd5d-0d0f98caffe1",
  "deckJsonHref": "/decks/019dd923-5e88-73ef-bd5d-0d0f98caffe1.json",
  "deckAnchorHref": "/decks/019dd923-5e88-73ef-bd5d-0d0f98caffe1#slide-17",
  "components": [
    {
      "bbox": null,
      "kind": "callout",
      "text": "Adaptive Parallel Reasoning (pictured) enables models to dynamically orchestrate branching inference through spawn() and join() operations, training both parent and child threads end-to-end using RL to optimize coordinated behavior.",
      "attrs": null,
      "subkind": null,
      "toolName": "Visual emphasis",
      "toolSlug": "visual-emphasis",
      "confidence": null,
      "componentId": "019dd952-41ac-7180-a81b-5d3c044e7f6d",
      "frameworkName": null,
      "frameworkSlug": null
    },
    {
      "bbox": {
        "h": 0.5,
        "w": 0.42,
        "x": 0.55,
        "y": 0.38
      },
      "kind": "image",
      "text": "Diagrams showing Adaptive Parallel Reasoning (APR) and Sample Set Aggregator (SSA) architectures.",
      "attrs": null,
      "subkind": "infographic",
      "toolName": null,
      "toolSlug": null,
      "confidence": null,
      "componentId": "03c27114-e653-494c-a23a-8a92fcd39665",
      "frameworkName": null,
      "frameworkSlug": null
    },
    {
      "bbox": {
        "h": 0.1,
        "w": 0.48,
        "x": 0.02,
        "y": 0.65
      },
      "kind": "list",
      "text": "Sample Set Aggregator (right) trains a compact model to fuse multiple reasoning samples into one coherent answer, outperforming naive re-ranking methods.",
      "attrs": null,
      "subkind": "bullet",
      "toolName": null,
      "toolSlug": null,
      "confidence": null,
      "componentId": "2dffaf3e-90d6-4f3d-b16a-056a2a570f6f",
      "frameworkName": null,
      "frameworkSlug": null
    },
    {
      "bbox": {
        "h": 0.25,
        "w": 0.48,
        "x": 0.02,
        "y": 0.38
      },
      "kind": "list",
      "text": "Adaptive Parallel Reasoning (pictured) enables models to dynamically orchestrate branching inference through spawn() and join() operations, training both parent and child threads end-to-end using RL to optimize coordinated behavior. This boosted performance on the Countdown task at 4K context: 83.4% (APR + RL) vs. 60.0% (baseline).",
      "attrs": null,
      "subkind": "bullet",
      "toolName": null,
      "toolSlug": null,
      "confidence": null,
      "componentId": "30b2bb0e-5e0b-4ec2-b846-d9f3d89039b1",
      "frameworkName": null,
      "frameworkSlug": null
    },
    {
      "bbox": {
        "h": 0.1,
        "w": 0.48,
        "x": 0.02,
        "y": 0.78
      },
      "kind": "list",
      "text": "Models like Gemini Deep Think, which shows its step-by-step reasoning transparently, exemplify this branch-and-evaluate paradigm in deployed systems.",
      "attrs": null,
      "subkind": "bullet",
      "toolName": null,
      "toolSlug": null,
      "confidence": null,
      "componentId": "39721486-f992-43b6-bd76-39b258f6c82a",
      "frameworkName": null,
      "frameworkSlug": null
    },
    {
      "bbox": null,
      "kind": "metric",
      "text": "Accuracy: 83.4%",
      "attrs": null,
      "subkind": "primary",
      "toolName": "Quantification",
      "toolSlug": "quantification",
      "confidence": null,
      "componentId": "019dd952-41ac-7180-a81b-60fe8767d2a1",
      "frameworkName": null,
      "frameworkSlug": null
    },
    {
      "bbox": {
        "h": 0.15,
        "w": 0.48,
        "x": 0.02,
        "y": 0.22
      },
      "kind": "paragraph",
      "text": "MoE routing scales capacity but preserves single-flow inference and doesn't change how the model thinks. A new route is branching multiple inference paths and aggregating them versus just going deeper or using wider models enables exploration, reduces hallucination, and better leverages parallel hardware.",
      "attrs": null,
      "subkind": "paragraph",
      "toolName": null,
      "toolSlug": null,
      "confidence": null,
      "componentId": "16fe5846-4e4c-4473-af12-93d91420fbfd",
      "frameworkName": null,
      "frameworkSlug": null
    },
    {
      "bbox": {
        "h": 0.05,
        "w": 0.6,
        "x": 0.02,
        "y": 0.14
      },
      "kind": "title",
      "text": "Parallel reasoning: beyond depth to branch-and-merge inference",
      "attrs": null,
      "subkind": "headline",
      "toolName": null,
      "toolSlug": null,
      "confidence": null,
      "componentId": "ab5bb6a8-bff2-45b4-8322-06826f5205b9",
      "frameworkName": null,
      "frameworkSlug": null
    }
  ],
  "metrics": [],
  "tools": [
    {
      "name": "Logical chain",
      "slug": "logical-chain",
      "agent": null,
      "layer": "slide",
      "matchId": "26e653c8-afb4-44fc-8bbd-d63872f6e90d",
      "evidence": "Adaptive Parallel Reasoning (pictured) enables models to dynamically orchestrate branching inference through spawn() and join() operations, training both parent and child threads end-to-end using RL to optimize",
      "confidence": 0.7
    }
  ],
  "frameworks": [],
  "arcBeats": [
    {
      "to": 189,
      "from": 12,
      "beatId": "019dd95a-0682-776c-8e35-9f380673831f",
      "arcName": "The Triple Take",
      "arcSlug": "triple-take",
      "beatName": "The Facts (What)",
      "beatSlug": "triple-take-the-facts-what",
      "evidence": "Sections 1-2 lay out research findings and industry data with charts and case studies.",
      "position": 1,
      "confidence": 70,
      "parentBeatName": "Setup",
      "parentBeatSlug": "setup"
    },
    {
      "to": 89,
      "from": 12,
      "beatId": "019dd95a-0682-776c-8e35-ac68557d958b",
      "arcName": "The Onion",
      "arcSlug": "onion",
      "beatName": "First Layer",
      "beatSlug": "onion-first-layer",
      "evidence": "Research section unpacks technical layer beneath the headlines.",
      "position": 2,
      "confidence": 45,
      "parentBeatName": "Development",
      "parentBeatSlug": "development"
    }
  ],
  "loops": [
    {
      "to": 30,
      "from": 13,
      "name": "Zoom In",
      "slug": "06-zoom-in",
      "bestFor": "Technical deep-dives, case studies, detailed analysis",
      "matchId": "019dd95a-07fe-70ce-8d3e-9252f6f92bd7",
      "evidence": "Big picture 'reasoning race' (p.13) zooms into o1, R1, DeepSeek, parallel reasoning, CoT mechanics.",
      "position": 2,
      "objective": "Drill from reasoning paradigm into specific reasoning research findings",
      "structure": "The Big Picture -> Key Area of Focus -> Specific Detail -> Implication",
      "confidence": 75,
      "description": "Start broad, then progressively focus on specific details that prove your point"
    }
  ],
  "imagePathAlt": null,
  "thumbSrc": null,
  "thumbSrcAlt": null,
  "locked": true
}