Jan Leike
jan-leikepersonPath: /knowledge-base/people/jan-leike/
E182Entity ID (EID)
Page Recorddatabase.json — merged from MDX frontmatter + Entity YAML + computed metrics at build time
{
"id": "jan-leike",
"numericId": null,
"path": "/knowledge-base/people/jan-leike/",
"filePath": "knowledge-base/people/jan-leike.mdx",
"title": "Jan Leike",
"quality": 27,
"readerImportance": 82,
"researchImportance": 39,
"tacticalValue": 82,
"contentFormat": "article",
"tractability": null,
"neglectedness": null,
"uncertainty": null,
"causalLevel": null,
"lastUpdated": "2026-03-13",
"dateCreated": "2026-02-15",
"llmSummary": "Biography of Jan Leike covering his career from Australian National University through DeepMind, OpenAI's Superalignment team, to his current role as head of the Alignment Science team at Anthropic. Documents his research on RLHF and scalable oversight, his May 2024 departure from OpenAI, and his current research priorities including weak-to-strong generalization and automated alignment techniques.",
"description": "Head of Alignment Science at Anthropic, formerly co-led OpenAI's Superalignment team",
"ratings": {
"novelty": 2,
"rigor": 3.5,
"actionability": 2,
"completeness": 5
},
"category": "people",
"subcategory": "safety-researchers",
"clusters": [
"ai-safety"
],
"metrics": {
"wordCount": 2585,
"tableCount": 1,
"diagramCount": 0,
"internalLinks": 46,
"externalLinks": 3,
"footnoteCount": 0,
"bulletRatio": 0.31,
"sectionCount": 23,
"hasOverview": true,
"structuralScore": 11
},
"suggestedQuality": 73,
"updateFrequency": 21,
"evergreen": true,
"wordCount": 2585,
"unconvertedLinks": [
{
"text": "jan.leike.name",
"url": "https://jan.leike.name/",
"resourceId": "2a84eb0982d4de6a",
"resourceTitle": "Personal website"
}
],
"unconvertedLinkCount": 1,
"convertedLinkCount": 0,
"backlinkCount": 26,
"hallucinationRisk": {
"level": "high",
"score": 90,
"factors": [
"biographical-claims",
"no-citations",
"low-rigor-score",
"low-quality-score"
]
},
"entityType": "person",
"redundancy": {
"maxSimilarity": 18,
"similarPages": [
{
"id": "ilya-sutskever",
"title": "Ilya Sutskever",
"path": "/knowledge-base/people/ilya-sutskever/",
"similarity": 18
},
{
"id": "chris-olah",
"title": "Chris Olah",
"path": "/knowledge-base/people/chris-olah/",
"similarity": 15
},
{
"id": "research-agendas",
"title": "AI Alignment Research Agenda Comparison",
"path": "/knowledge-base/responses/research-agendas/",
"similarity": 15
},
{
"id": "technical-research",
"title": "Technical AI Safety Research",
"path": "/knowledge-base/responses/technical-research/",
"similarity": 15
},
{
"id": "ssi",
"title": "Safe Superintelligence Inc (SSI)",
"path": "/knowledge-base/organizations/ssi/",
"similarity": 14
}
]
},
"changeHistory": [
{
"date": "2026-02-23",
"branch": "feat/batch-improve-high-risk-pages",
"title": "Auto-improve (standard): Jan Leike",
"summary": "Improved \"Jan Leike\" via standard pipeline (1147.1s). Quality score: 88. Issues resolved: Footnote [^19] cites Sam Altman's X post but the URL points ; The <F e=\"openai\" f=\"\">20%</F> component tag in the OpenAI s; Bare URL 'aligned.substack.com' appears in the Public Commun.",
"duration": "1147.1s",
"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"
},
{
"date": "2026-02-17",
"branch": "claude/review-wiki-editing-scCul",
"title": "Wiki editing system refactoring",
"summary": "Six refactors to the wiki editing pipeline: (1) extracted shared regex patterns to `crux/lib/patterns.ts`, (2) refactored validation in page-improver to use in-process engine calls instead of subprocess spawning, (3) split the 694-line `phases.ts` into 7 individual phase modules under `phases/`, (4) created shared LLM abstraction `crux/lib/llm.ts` unifying duplicated streaming/retry/tool-loop code, (5) added Zod schemas for LLM JSON response validation, (6) decomposed 820-line mermaid validation into `crux/lib/mermaid-checks.ts` (604 lines) + slim orchestrator (281 lines). Follow-up review integrated patterns.ts across 19+ files, fixed dead imports, corrected ToolHandler type, wired mdx-utils.ts to use shared patterns, replaced hardcoded model strings with MODELS constants, replaced `new Anthropic()` with `createLlmClient()`, replaced inline `extractText` implementations with shared `extractText()` from llm.ts, integrated `MARKDOWN_LINK_RE` into link validators, added `objectivityIssues` to the `AnalysisResult` type (removing an unsafe cast in utils.ts), fixed CI failure from eager client creation, and tested the full pipeline by improving 3 wiki pages. After manual review of 3 improved pages, fixed 8 systematic pipeline issues: (1) added content preservation instructions to prevent polish-tier content loss, (2) made auto-grading default after --apply, (3) added polish-tier citation suppression to prevent fabricated citations, (4) added Quick Assessment table requirement for person pages, (5) added required Overview section enforcement, (6) added section deduplication and content repetition checks to review phase, (7) added bare URL→markdown link conversion instruction, (8) extended biographical claim checker to catch publication/co-authorship and citation count claims.\n\nSubsequent iterative testing and prompt refinement: ran pipeline on jan-leike, chris-olah, far-ai pages. Discovered and fixed: (a) `<!-- NEEDS CITATION -->` HTML comments break MDX compilation (changed to `{/* NEEDS CITATION */}`), (b) excessive citation markers at polish tier — added instruction to only mark NEW claims (max 3-5 per page), (c) editorial meta-comments cluttering output — added no-meta-comments instruction, (d) thin padding sections — added anti-padding instruction, (e) section deduplication needed stronger emphasis — added merge instruction with common patterns. Final test results: jan-leike 1254→1997 words, chris-olah 1187→1687 words, far-ai 1519→2783 words, miri-era 2678→4338 words; all MDX compile, zero critical issues.",
"pr": 184
}
],
"coverage": {
"passing": 6,
"total": 13,
"targets": {
"tables": 10,
"diagrams": 1,
"internalLinks": 21,
"externalLinks": 13,
"footnotes": 8,
"references": 8
},
"actuals": {
"tables": 1,
"diagrams": 0,
"internalLinks": 46,
"externalLinks": 3,
"footnotes": 0,
"references": 3,
"quotesWithQuotes": 0,
"quotesTotal": 0,
"accuracyChecked": 0,
"accuracyTotal": 0
},
"items": {
"llmSummary": "green",
"schedule": "green",
"entity": "green",
"editHistory": "green",
"overview": "green",
"tables": "amber",
"diagrams": "red",
"internalLinks": "green",
"externalLinks": "amber",
"footnotes": "red",
"references": "amber",
"quotes": "red",
"accuracy": "red"
},
"editHistoryCount": 3,
"ratingsString": "N:2 R:3.5 A:2 C:5"
},
"readerRank": 74,
"researchRank": 351,
"recommendedScore": 116.77
}External Links
{
"wikidata": "https://www.wikidata.org/wiki/Q123130693"
}Backlinks (26)
| id | title | type | relationship |
|---|---|---|---|
| anthropic | Anthropic | organization | research |
| openai | OpenAI | organization | research |
| dario-amodei | Dario Amodei | person | — |
| ilya-sutskever | Ilya Sutskever | person | — |
| paul-christiano | Paul Christiano | person | — |
| accident-risks | AI Accident Risk Cruxes | crux | — |
| why-alignment-easy | Why Alignment Might Be Easy | argument | — |
| safety-capability-tradeoff | Safety-Capability Tradeoff Model | analysis | — |
| anthropic-valuation | Anthropic Valuation Analysis | analysis | — |
| chai | CHAI (Center for Human-Compatible AI) | organization | — |
| frontier-ai-comparison | Frontier AI Company Comparison (2026) | concept | — |
| daniela-amodei | Daniela Amodei | person | — |
| evan-hubinger | Evan Hubinger | person | — |
| helen-toner | Helen Toner | person | — |
| __index__/knowledge-base/people | People | concept | — |
| leopold-aschenbrenner | Leopold Aschenbrenner | person | — |
| sam-altman | Sam Altman | person | — |
| ai-assisted | AI-Assisted Alignment | approach | — |
| alignment | AI Alignment | approach | — |
| anthropic-core-views | Anthropic Core Views | safety-agenda | — |
| corporate-influence | Corporate Influence on AI Policy | crux | — |
| lab-culture | AI Lab Safety Culture | approach | — |
| research-agendas | AI Alignment Research Agenda Comparison | crux | — |
| scalable-oversight | Scalable Oversight | safety-agenda | — |
| optimistic | Optimistic Alignment Worldview | concept | — |
| longtermwiki-value-proposition | LongtermWiki Value Proposition | concept | — |