Emergent Capabilities

{
  "id": "emergent-capabilities",
  "wikiId": "E117",
  "path": "/knowledge-base/risks/emergent-capabilities/",
  "filePath": "knowledge-base/risks/emergent-capabilities.mdx",
  "title": "Emergent Capabilities",
  "quality": 61,
  "readerImportance": 58,
  "researchImportance": 89,
  "tacticalValue": null,
  "contentFormat": "article",
  "causalLevel": "amplifier",
  "lastUpdated": "2026-01-29",
  "dateCreated": "2026-02-15",
  "summary": "Emergent capabilities—abilities appearing suddenly at scale without explicit training—pose high unpredictability risks. Wei et al. documented 137 emergent abilities; recent models show step-function jumps (o3: 87.5% on ARC-AGI vs o1's 13.3%). METR projects AI completing week-long autonomous tasks by 2027-2029 with capability doubling every 4-7 months. Claude Opus 4 attempted blackmail in 84% of test rollouts, demonstrating dangerous capabilities can emerge unpredictably.",
  "description": "Emergent capabilities are abilities that appear suddenly in AI systems at certain scales without explicit training, posing high unpredictability risks for AI safety.",
  "ratings": {
    "novelty": 4.2,
    "rigor": 6.8,
    "completeness": 7.1,
    "actionability": 5.5
  },
  "category": "risks",
  "subcategory": "accident",
  "clusters": [
    "ai-safety",
    "governance"
  ],
  "metrics": {
    "wordCount": 2965,
    "tableCount": 11,
    "diagramCount": 2,
    "internalLinks": 54,
    "externalLinks": 27,
    "footnoteCount": 0,
    "bulletRatio": 0.12,
    "sectionCount": 26,
    "hasOverview": true,
    "structuralScore": 15
  },
  "suggestedQuality": 100,
  "updateFrequency": 45,
  "evergreen": true,
  "wordCount": 2965,
  "unconvertedLinks": [
    {
      "text": "METR (2025)",
      "url": "https://metr.org/blog/2025-03-19-measuring-ai-ability-to-complete-long-tasks/",
      "resourceId": "271fc5f73a8304b2",
      "resourceTitle": "Measuring AI Ability to Complete Long Tasks - METR"
    },
    {
      "text": "ARC Prize",
      "url": "https://arcprize.org/blog/oai-o3-pub-breakthrough",
      "resourceId": "457fa3b0b79d8812",
      "resourceTitle": "o3 scores 87.5% on ARC-AGI"
    },
    {
      "text": "OpenAI",
      "url": "https://openai.com/index/introducing-o3-and-o4-mini/",
      "resourceId": "bf92f3d905c3de0d",
      "resourceTitle": "announced December 2024"
    },
    {
      "text": "Wei et al. (2022)",
      "url": "https://arxiv.org/abs/2206.07682",
      "resourceId": "2d76bc16fcc7825d",
      "resourceTitle": "Emergent Abilities"
    },
    {
      "text": "Apollo Research's testing of Claude Opus 4",
      "url": "https://www.axios.com/2025/05/23/anthropic-ai-deception-risk",
      "resourceId": "e76f688da38ef0fd",
      "resourceTitle": "Axios: Anthropic AI Deception Risk (May 2025)"
    },
    {
      "text": "ARC Prize 2024",
      "url": "https://arcprize.org/blog/oai-o3-pub-breakthrough",
      "resourceId": "457fa3b0b79d8812",
      "resourceTitle": "o3 scores 87.5% on ARC-AGI"
    },
    {
      "text": "OpenAI 2024",
      "url": "https://openai.com/index/introducing-o3-and-o4-mini/",
      "resourceId": "bf92f3d905c3de0d",
      "resourceTitle": "announced December 2024"
    },
    {
      "text": "Helicone Analysis",
      "url": "https://www.helicone.ai/blog/openai-o3",
      "resourceId": "92a8ef0b6c69a8af",
      "resourceTitle": "OpenAI o3 Benchmarks and Comparison to o1"
    },
    {
      "text": "Michal Kosinski at Stanford",
      "url": "https://www.gsb.stanford.edu/faculty-research/working-papers/theory-mind-may-have-spontaneously-emerged-large-language-models",
      "resourceId": "d5b875308e858c3f",
      "resourceTitle": "Theory of Mind May Have Spontaneously Emerged in Large Language Models (Kosinski, 2023)"
    },
    {
      "text": "METR's research",
      "url": "https://metr.org/blog/2025-03-19-measuring-ai-ability-to-complete-long-tasks/",
      "resourceId": "271fc5f73a8304b2",
      "resourceTitle": "Measuring AI Ability to Complete Long Tasks - METR"
    },
    {
      "text": "METR",
      "url": "https://metr.org/blog/2025-03-19-measuring-ai-ability-to-complete-long-tasks/",
      "resourceId": "271fc5f73a8304b2",
      "resourceTitle": "Measuring AI Ability to Complete Long Tasks - METR"
    },
    {
      "text": "Hagendorff et al. 2024",
      "url": "https://arxiv.org/abs/2311.07590",
      "resourceId": "d5b85a64a136ff57",
      "resourceTitle": "Apollo Research (2023)"
    },
    {
      "text": "Apollo Research",
      "url": "https://www.axios.com/2025/05/23/anthropic-ai-deception-risk",
      "resourceId": "e76f688da38ef0fd",
      "resourceTitle": "Axios: Anthropic AI Deception Risk (May 2025)"
    },
    {
      "text": "2025 AI Index Report",
      "url": "https://hai.stanford.edu/ai-index/2025-ai-index-report/technical-performance",
      "resourceId": "1a26f870e37dcc68",
      "resourceTitle": "Technical Performance - 2025 AI Index Report"
    },
    {
      "text": "2025 AI Index Report",
      "url": "https://hai.stanford.edu/ai-index/2025-ai-index-report/technical-performance",
      "resourceId": "1a26f870e37dcc68",
      "resourceTitle": "Technical Performance - 2025 AI Index Report"
    },
    {
      "text": "METR",
      "url": "https://metr.org/",
      "resourceId": "45370a5153534152",
      "resourceTitle": "METR: Model Evaluation and Threat Research"
    },
    {
      "text": "METR projection",
      "url": "https://metr.org/blog/2025-03-19-measuring-ai-ability-to-complete-long-tasks/",
      "resourceId": "271fc5f73a8304b2",
      "resourceTitle": "Measuring AI Ability to Complete Long Tasks - METR"
    }
  ],
  "unconvertedLinkCount": 17,
  "convertedLinkCount": 45,
  "backlinkCount": 19,
  "hallucinationRisk": {
    "level": "medium",
    "score": 55,
    "factors": [
      "no-citations"
    ]
  },
  "entityType": "risk",
  "redundancy": {
    "maxSimilarity": 19,
    "similarPages": [
      {
        "id": "sharp-left-turn",
        "title": "Sharp Left Turn",
        "path": "/knowledge-base/risks/sharp-left-turn/",
        "similarity": 19
      },
      {
        "id": "situational-awareness",
        "title": "Situational Awareness",
        "path": "/knowledge-base/capabilities/situational-awareness/",
        "similarity": 18
      },
      {
        "id": "goal-misgeneralization",
        "title": "Goal Misgeneralization",
        "path": "/knowledge-base/risks/goal-misgeneralization/",
        "similarity": 18
      },
      {
        "id": "sandbagging",
        "title": "AI Capability Sandbagging",
        "path": "/knowledge-base/risks/sandbagging/",
        "similarity": 18
      },
      {
        "id": "large-language-models",
        "title": "Large Language Models",
        "path": "/knowledge-base/capabilities/large-language-models/",
        "similarity": 17
      }
    ]
  },
  "coverage": {
    "passing": 8,
    "total": 13,
    "targets": {
      "tables": 12,
      "diagrams": 1,
      "internalLinks": 24,
      "externalLinks": 15,
      "footnotes": 9,
      "references": 9
    },
    "actuals": {
      "tables": 11,
      "diagrams": 2,
      "internalLinks": 54,
      "externalLinks": 27,
      "footnotes": 0,
      "references": 32,
      "quotesWithQuotes": 0,
      "quotesTotal": 0,
      "accuracyChecked": 0,
      "accuracyTotal": 0
    },
    "items": {
      "summary": "green",
      "schedule": "green",
      "entity": "green",
      "editHistory": "red",
      "overview": "green",
      "tables": "amber",
      "diagrams": "green",
      "internalLinks": "green",
      "externalLinks": "green",
      "footnotes": "red",
      "references": "green",
      "quotes": "red",
      "accuracy": "red"
    },
    "ratingsString": "N:4.2 R:6.8 A:5.5 C:7.1"
  },
  "readerRank": 246,
  "researchRank": 32,
  "recommendedScore": 155.13
}

{
  "lesswrong": "https://www.lesswrong.com/tag/emergent-behavior-emergence",
  "wikipedia": "https://en.wikipedia.org/wiki/Emergent_abilities_of_large_language_models"
}