Large Language Models

{
  "id": "language-models",
  "numericId": null,
  "path": "/knowledge-base/capabilities/language-models/",
  "filePath": "knowledge-base/capabilities/language-models.mdx",
  "title": "Large Language Models",
  "quality": 60,
  "readerImportance": 94,
  "researchImportance": 76.5,
  "tacticalValue": 74,
  "contentFormat": "article",
  "tractability": null,
  "neglectedness": null,
  "uncertainty": null,
  "causalLevel": null,
  "lastUpdated": "2026-03-13",
  "dateCreated": "2026-02-15",
  "llmSummary": "Comprehensive analysis of LLM capabilities showing rapid progress from GPT-2 (1.5B parameters, 2019) to GPT-5 and Gemini 2.5 (2025), with training costs growing 2.4x annually and projected to exceed \\$1B by 2027. Documents emergence of inference-time scaling paradigm, mechanistic interpretability advances including Gemma Scope 2, multilingual alignment research, factuality benchmarking via FACTS suite, and identifies key safety concerns including 8-45% hallucination rates, persuasion capabilities, and growing autonomous agent capabilities.",
  "description": "Foundation models trained on text that demonstrate emergent capabilities and represent the primary driver of current AI capabilities and risks, with rapid progression from GPT-2 (1.5B parameters, 2019) to GPT-5 and Gemini 2.5 (2025) showing predictable scaling laws alongside unpredictable capability emergence",
  "ratings": {
    "novelty": 4.5,
    "rigor": 6.5,
    "actionability": 5,
    "completeness": 7.5
  },
  "category": "capabilities",
  "subcategory": "core",
  "clusters": [
    "ai-safety",
    "governance"
  ],
  "metrics": {
    "wordCount": 7369,
    "tableCount": 17,
    "diagramCount": 1,
    "internalLinks": 60,
    "externalLinks": 67,
    "footnoteCount": 1,
    "bulletRatio": 0.17,
    "sectionCount": 55,
    "hasOverview": true,
    "structuralScore": 15
  },
  "suggestedQuality": 100,
  "updateFrequency": 21,
  "evergreen": true,
  "wordCount": 7369,
  "unconvertedLinks": [
    {
      "text": "Summarizing Books with Human Feedback (Wu et al., 2021)",
      "url": "https://arxiv.org/abs/2109.10862",
      "resourceId": "54eec19853953598",
      "resourceTitle": "[2109.10862] Recursively Summarizing Books with Human Feedback"
    },
    {
      "text": "TruthfulQA: Measuring How Models Mimic Human Falsehoods (Lin et al., 2022)",
      "url": "https://arxiv.org/abs/2109.07958",
      "resourceId": "fe2a3307a3dae3e5",
      "resourceTitle": "Kenton et al. (2021)"
    },
    {
      "text": "WebGPT: Browser-Assisted Question-Answering with Human Feedback (Nakano et al., 2021)",
      "url": "https://arxiv.org/abs/2112.09332",
      "resourceId": "3225de3850d36a20",
      "resourceTitle": "OpenAI WebGPT behavior"
    },
    {
      "text": "Evaluating Large Language Models Trained on Code (Chen et al., 2021)",
      "url": "https://arxiv.org/abs/2107.03374",
      "resourceId": "176fdaf24fa29d4c",
      "resourceTitle": "Evaluating Large Language Models Trained on Code"
    },
    {
      "text": "Helpful to a Fault (2025)",
      "url": "https://arxiv.org/abs/2502.09933",
      "resourceId": "92dfabf49665e538",
      "resourceTitle": "Helpful to a Fault (2025)."
    },
    {
      "text": "FACTS Grounding",
      "url": "https://arxiv.org/abs/2501.03200",
      "resourceId": "1f7d271db2f8a756",
      "resourceTitle": "Google DeepMind, \"FACTS Grounding: A new benchmark for evaluating the factuality of large language models\" (https://a..."
    },
    {
      "text": "Gemini 2.0",
      "url": "https://blog.google/technology/google-deepmind/google-gemini-ai-update-december-2024/",
      "resourceId": "08aca1a4de71818f",
      "resourceTitle": "Gemini 2.0 Flash"
    },
    {
      "text": "GPT-5 System Card",
      "url": "https://openai.com/index/gpt-5-system-card/",
      "resourceId": "817c3cbf13144f20",
      "resourceTitle": "OpenAI."
    },
    {
      "text": "Hello GPT-4o (2024)",
      "url": "https://openai.com/index/hello-gpt-4o/",
      "resourceId": "ee605bab036068f0",
      "resourceTitle": "GPT-4o"
    },
    {
      "text": "GPT-5 System Card",
      "url": "https://openai.com/index/gpt-5-system-card/",
      "resourceId": "817c3cbf13144f20",
      "resourceTitle": "OpenAI."
    },
    {
      "text": "Llama 4 Herd",
      "url": "https://ai.meta.com/blog/llama-4-multimodal-intelligence/",
      "resourceId": "05f285e9757b863c",
      "resourceTitle": "LlamaFirewall"
    },
    {
      "text": "TruthfulQA",
      "url": "https://arxiv.org/abs/2109.07958",
      "resourceId": "fe2a3307a3dae3e5",
      "resourceTitle": "Kenton et al. (2021)"
    },
    {
      "text": "FACTS Grounding",
      "url": "https://arxiv.org/abs/2501.03200",
      "resourceId": "1f7d271db2f8a756",
      "resourceTitle": "Google DeepMind, \"FACTS Grounding: A new benchmark for evaluating the factuality of large language models\" (https://a..."
    },
    {
      "text": "Vectara",
      "url": "https://github.com/vectara/hallucination-leaderboard",
      "resourceId": "b44f883dc65dd0e9",
      "resourceTitle": "Vectara, \"Hallucination Leaderboard\" (https://github.com/vectara/hallucination-leaderboard)"
    },
    {
      "text": "HalluLens (ACL 2025)",
      "url": "https://aclanthology.org/2025.acl-long.1176/",
      "resourceId": "351b4e7354e2dc5b",
      "resourceTitle": "HalluLens, ACL 2025 (https://aclanthology.org/2025.acl-long.1176/)"
    },
    {
      "text": "CheckIfExist (2025)",
      "url": "https://arxiv.org/abs/2502.09802",
      "resourceId": "3df9ab6bab412a71",
      "resourceTitle": "CheckIfExist authors, \"CheckIfExist: Citation Hallucination Detection in RAG Systems\" (https://arxiv.org/abs/2502.09802)"
    },
    {
      "text": "CheckIfExist",
      "url": "https://arxiv.org/abs/2502.09802",
      "resourceId": "3df9ab6bab412a71",
      "resourceTitle": "CheckIfExist authors, \"CheckIfExist: Citation Hallucination Detection in RAG Systems\" (https://arxiv.org/abs/2502.09802)"
    }
  ],
  "unconvertedLinkCount": 17,
  "convertedLinkCount": 7,
  "backlinkCount": 52,
  "hallucinationRisk": {
    "level": "high",
    "score": 80,
    "factors": [
      "low-citation-density",
      "severe-truncation"
    ],
    "integrityIssues": [
      "severe-truncation"
    ]
  },
  "entityType": "capability",
  "redundancy": {
    "maxSimilarity": 22,
    "similarPages": [
      {
        "id": "agentic-ai",
        "title": "Agentic AI",
        "path": "/knowledge-base/capabilities/agentic-ai/",
        "similarity": 22
      },
      {
        "id": "reasoning",
        "title": "Reasoning and Planning",
        "path": "/knowledge-base/capabilities/reasoning/",
        "similarity": 22
      },
      {
        "id": "solutions",
        "title": "AI Safety Solution Cruxes",
        "path": "/knowledge-base/cruxes/solutions/",
        "similarity": 22
      },
      {
        "id": "why-alignment-hard",
        "title": "Why Alignment Might Be Hard",
        "path": "/knowledge-base/debates/why-alignment-hard/",
        "similarity": 22
      },
      {
        "id": "deep-learning-era",
        "title": "Deep Learning Revolution (2012-2020)",
        "path": "/knowledge-base/history/deep-learning-era/",
        "similarity": 21
      }
    ]
  },
  "changeHistory": [
    {
      "date": "2026-03-13",
      "branch": "auto-update/2026-03-13",
      "title": "Auto-improve (standard): Large Language Models",
      "summary": "Improved \"Large Language Models\" via standard pipeline (489.5s). Quality score: 71. Issues resolved: Page is truncated — the section 'Cybersecurity Capabilities ; Frontmatter 'lastEdited' is '2026-03-13', which is a future ; Footnote references such as [^rc-ed93], [^rc-e8f4], [^rc-c58.",
      "duration": "489.5s",
      "cost": "$5-8"
    },
    {
      "date": "2026-03-11",
      "branch": "auto-update/2026-03-11",
      "title": "Auto-improve (standard): Large Language Models",
      "summary": "Improved \"Large Language Models\" via standard pipeline (515.5s). Quality score: 72. Issues resolved: Frontmatter 'lastEdited' is set to '2026-03-11', which is a ; Capability Progression Timeline table contains a row for 'Cl; The 'FACTS Benchmark Suite' row in the Hallucination benchma.",
      "duration": "515.5s",
      "cost": "$5-8"
    },
    {
      "date": "2026-03-10",
      "branch": "auto-update/2026-03-10",
      "title": "Auto-improve (standard): Large Language Models",
      "summary": "Improved \"Large Language Models\" via standard pipeline (516.9s). Quality score: 71. Issues resolved: Frontmatter 'lastEdited' date is '2026-03-10', which is a fu; Capability Progression Timeline: '<EntityLink id=\"E1030\">Cla; Benchmark Performance Comparison table: TruthfulQA row has '.",
      "duration": "516.9s",
      "cost": "$5-8"
    },
    {
      "date": "2026-03-09",
      "branch": "auto-update/2026-03-09",
      "title": "Auto-improve (standard): Large Language Models",
      "summary": "Improved \"Large Language Models\" via standard pipeline (510.4s). Quality score: 71. Issues resolved: Page is truncated mid-sentence at the end: '...achieving com; Frontmatter field 'lastEdited' is dated '2026-03-09' which i; Capability Progression Timeline table contains a duplicate E.",
      "duration": "510.4s",
      "cost": "$5-8"
    },
    {
      "date": "2026-03-07",
      "branch": "auto-update/2026-03-07",
      "title": "Auto-improve (standard): Large Language Models",
      "summary": "Improved \"Large Language Models\" via standard pipeline (1306.4s). Quality score: 71. Issues resolved: Frontmatter: 'llmSummary' contains an escaped dollar sign (\\; Capability Progression Timeline table: 'Claude Opus 4.5' row; Benchmark Performance Comparison table: ARC-AGI-2 row shows .",
      "duration": "1306.4s",
      "cost": "$5-8"
    },
    {
      "date": "2026-02-18",
      "branch": "claude/fix-issue-240-N5irU",
      "title": "Surface tacticalValue in /wiki table and score 53 pages",
      "summary": "Added `tacticalValue` to `ExploreItem` interface, `getExploreItems()` mappings, the `/wiki` explore table (new sortable \"Tact.\" column), and the card view sort dropdown. Scored 49 new pages with tactical values (4 were already scored), bringing total to 53.",
      "model": "sonnet-4",
      "duration": "~30min"
    }
  ],
  "coverage": {
    "passing": 8,
    "total": 13,
    "targets": {
      "tables": 29,
      "diagrams": 3,
      "internalLinks": 59,
      "externalLinks": 37,
      "footnotes": 22,
      "references": 22
    },
    "actuals": {
      "tables": 17,
      "diagrams": 1,
      "internalLinks": 60,
      "externalLinks": 67,
      "footnotes": 1,
      "references": 36,
      "quotesWithQuotes": 0,
      "quotesTotal": 0,
      "accuracyChecked": 0,
      "accuracyTotal": 0
    },
    "items": {
      "llmSummary": "green",
      "schedule": "green",
      "entity": "green",
      "editHistory": "green",
      "overview": "green",
      "tables": "amber",
      "diagrams": "amber",
      "internalLinks": "green",
      "externalLinks": "green",
      "footnotes": "amber",
      "references": "green",
      "quotes": "red",
      "accuracy": "red"
    },
    "editHistoryCount": 6,
    "ratingsString": "N:4.5 R:6.5 A:5 C:7.5"
  },
  "readerRank": 4,
  "researchRank": 109,
  "recommendedScore": 188.86
}

{
  "wikipedia": "https://en.wikipedia.org/wiki/Large_language_model",
  "lesswrong": "https://www.lesswrong.com/tag/language-models-llms",
  "eaForum": "https://forum.effectivealtruism.org/topics/large-language-models",
  "grokipedia": "https://grokipedia.com/page/Large_language_model"
}