{"id":"48b9bfa6f6e2078f4e17b78c944aa24a2c67228f29ec80b9f63d329765b64bd9","title":"Jack Clark argues in Import AI that AI research taste could bring recursive self-improvement closer","argument":"Import AI was published on October 12, 2026. Jack Clark argues that AI systems are getting better at what researchers call “good scientific taste”: choosing problems, designing experiments and interpreting results. He describes TasteVal, a benchmark from P-Zero Research that tests how models iteratively design experiments and draw conclusions for fixed research problems. Clark reports that its strongest model exceeded the human baseline on the benchmark, at lower cost per attempt. He connects this result to recursive self-improvement, reasoning that creativity may affect how quickly AI systems can develop the capacity to improve themselves. Clark says he now thinks full recursive self-improvement is more likely in 2027 than he had thought before, when he considered it more likely in 2028. This is his forecast, not a reported outcome. TasteVal measures an experimental component of research taste; the account does not establish that benchmark performance will translate into systems improving themselves. The newsletter also discusses visual reasoning benchmarks, space, open-weight models, government computing, and a fictional future in which direct conversations with powerful synthetic minds pose risks to people.","topicIds":["artificial-general-intelligence"],"domain":"other","claimType":"attributed reading","limits":"Text as retrieved on October 23, 2026; originally published October 12, 2026. TasteVal measures an experimental component of research taste for a fixed research problem; the newsletter does not establish that benchmark performance will translate into recursive self-improvement.","speaker":{"name":"Jack Clark","roleAtSource":"Publisher"},"source":{"title":"Import AI 476: Creativity and RSI; worse-than-China open weight models; AI goes to space","publisher":"Jack Clark","format":"Article","url":"https://jack-clark.net/2026/10/12/import-ai-476-creativity-and-rsi-worse-than-china-open-weight-models-ai-goes-to-space/","publishedAt":"2026-10-12T12:30:39.000Z","dateBasis":"Date supplied by the publisher’s feed."},"evidence":[{"text":"Jack Clark reports that TasteVal measures how well a model iteratively designs experiments and draws conclusions from their outcomes for a fixed research problem.","excerpt":"given a fixed research problem, we measure how well a model iteratively designs experiments and draws conclusions from their outcomes","url":"https://jack-clark.net/2026/10/12/import-ai-476-creativity-and-rsi-worse-than-china-open-weight-models-ai-goes-to-space/","relationship":"Source-stated"},{"text":"Clark argues that creativity may be important to how quickly AI systems develop the capacity to improve themselves.","excerpt":"creativity seemed like a key variable for how fast AI systems could develop the capacity to improve themselves.","url":"https://jack-clark.net/2026/10/12/import-ai-476-creativity-and-rsi-worse-than-china-open-weight-models-ai-goes-to-space/","relationship":"Source-stated"},{"text":"Clark says he now considers full recursive self-improvement more likely in 2027 than he did previously; this is his forecast.","excerpt":"I now think it’s more likely we’ll see full RSI become possible in 2027, whereas before I thought it was more likely in 2028.","url":"https://jack-clark.net/2026/10/12/import-ai-476-creativity-and-rsi-worse-than-china-open-weight-models-ai-goes-to-space/","relationship":"Source-stated"}],"sourceMedia":{"status":"unchecked","sourceUrl":"https://jack-clark.net/2026/10/12/import-ai-476-creativity-and-rsi-worse-than-china-open-weight-models-ai-goes-to-space/","checkedAt":null,"items":[]},"assumptions":[],"openQuestion":{"attribution":"Open question","text":"Will improved performance on TasteVal translate into AI systems becoming able to improve themselves?"},"whatWouldChangeView":{"attribution":"Change condition not recorded","text":""},"editorialComparisons":[],"live":{"title":"Jack Clark argues in Import AI that AI research taste could bring recursive self-improvement closer","summary":"Import AI was published on October 12, 2026. Jack Clark argues that AI systems are getting better at what researchers call “good scientific taste”: choosing problems, designing experiments and interpreting results. He describes TasteVal, a benchmark from P-Zero Research that tests how models iteratively design experiments and draw conclusions for fixed research problems. Clark reports that its strongest model exceeded the human baseline on the benchmark, at lower cost per attempt. He connects this result to recursive self-improvement, reasoning that creativity may affect how quickly AI systems can develop the capacity to improve themselves. Clark says he now thinks full recursive self-improvement is more likely in 2027 than he had thought before, when he considered it more likely in 2028. This is his forecast, not a reported outcome. TasteVal measures an experimental component of research taste; the account does not establish that benchmark performance will translate into systems improving themselves. The newsletter also discusses visual reasoning benchmarks, space, open-weight models, government computing, and a fictional future in which direct conversations with powerful synthetic minds pose risks to people.","question":"Will improved performance on TasteVal translate into AI systems becoming able to improve themselves?","claims":[{"text":"Jack Clark reports that TasteVal measures how well a model iteratively designs experiments and draws conclusions from their outcomes for a fixed research problem.","excerpt":"given a fixed research problem, we measure how well a model iteratively designs experiments and draws conclusions from their outcomes","support":"source-stated"},{"text":"Clark argues that creativity may be important to how quickly AI systems develop the capacity to improve themselves.","excerpt":"creativity seemed like a key variable for how fast AI systems could develop the capacity to improve themselves.","support":"source-stated"},{"text":"Clark says he now considers full recursive self-improvement more likely in 2027 than he did previously; this is his forecast.","excerpt":"I now think it’s more likely we’ll see full RSI become possible in 2027, whereas before I thought it was more likely in 2028.","support":"source-stated"}],"limitations":["Text as retrieved on October 23, 2026; originally published October 12, 2026.","TasteVal measures an experimental component of research taste for a fixed research problem; the newsletter does not establish that benchmark performance will translate into recursive self-improvement."],"mediaId":null,"correctionReason":null,"id":"48b9bfa6f6e2078f4e17b78c944aa24a2c67228f29ec80b9f63d329765b64bd9","publicationId":"aa4be3ed-34a0-abd5-88b5-53e872ad4808","version":1,"sourceUrl":"https://jack-clark.net/2026/10/12/import-ai-476-creativity-and-rsi-worse-than-china-open-weight-models-ai-goes-to-space/","sourceTitle":"Import AI 476: Creativity and RSI; worse-than-China open weight models; AI goes to space","publisher":"Jack Clark","publishedAt":"2026-10-12T12:30:39.000Z","originalPublished":{"value":"2026-10-12T12:30:39.000Z","precision":"instant","basis":"backfill:feed_published"},"firstObservedAt":"2026-10-23T20:01:48.876Z","createdAt":"2026-10-31T11:34:28.474Z","historical":true,"entityIds":["jack-clark"],"topicIds":["artificial-general-intelligence"],"media":[],"sourceMedia":{"status":"unchecked","sourceUrl":"https://jack-clark.net/2026/10/12/import-ai-476-creativity-and-rsi-worse-than-china-open-weight-models-ai-goes-to-space/","checkedAt":null,"items":[]},"editorialDecision":{"outcome":"accept","mediaId":null,"inputHash":"14af1025635d34a870dbd0ba3c157d4c96b30569df438b70d12eeef756ac479c","rubricVersion":"unrecorded","provider":"unrecorded","model":"anthropic/claude-opus-5.5","reasonCodes":["bridge_admitted_by_model","second_opinion_categorical"],"confidence":1,"decidedAt":"2026-10-23T20:01:46.020Z"},"provenance":{"textHash":"cb3dd2188128c6892c8012e3c6b03e359737ac196638ff9889895bbc9d7e4a97","observationId":"7d1dafa435e87dc1b5bba98375af4282c718eea23d5be2bc60ed499c9973ddb3","retrievedAt":"2026-10-23T20:03:37.266Z","kind":"publisher-page","preparer":{"model":"openai/gpt-6-luna","runId":"425a5e3f-9304-029e-dbbc-71514e0257ab"},"reviewer":{"model":"anthropic/claude-opus-5.5","runId":"39c956c8-6df2-5248-b9de-9713bc79377a"}},"audit":{"status":"passed","reviewedAt":"2026-10-31T11:34:28.165Z","model":"anthropic/claude-opus-5.5"}},"canonicalUrl":"https://escapemeridian.com/readings/jack-clark-argues-in-import-ai-that-ai-research-taste-could-48b9bfa6","formats":{"html":"/readings/jack-clark-argues-in-import-ai-that-ai-research-taste-could-48b9bfa6","markdown":"/readings/jack-clark-argues-in-import-ai-that-ai-research-taste-could-48b9bfa6.md","json":"/readings/jack-clark-argues-in-import-ai-that-ai-research-taste-could-48b9bfa6.json"},"attribution":"Headline and summary by Escape Meridian, prepared by Luna from the attributed source. A source check is not independent confirmation of the source’s claims.","readerContext":{"version":"reader-context-v1","attribution":"Reference context selected by Escape Meridian; separate from source evidence. Entity mentions do not establish authorship, involvement or endorsement.","terms":[{"id":"term-recursive-self-improvement","name":"Recursive self-improvement","explanation":"An AI system improving its own capabilities, such as its code, training methods or reasoning, then using the gains to improve further. Each cycle could hypothetically produce a more capable system.","referenceUrl":"https://hai.stanford.edu/what-is-recursive-self-improvement"},{"id":"term-benchmark","name":"Benchmark","explanation":"A defined set of tests used to compare systems. Results depend on the tasks and measures selected.","referenceUrl":"https://arxiv.org/abs/2211.09110"},{"id":"term-open-weights","name":"Open weights","explanation":"A model whose trained numbers are published so others can download, run and inspect it; the license and training data may still be restricted.","referenceUrl":"https://opensource.org/ai/open-weights"}],"entities":[{"id":"jack-clark","name":"Jack Clark","kind":"person","context":"Jack Clark writes Import AI, a newsletter covering AI research, development and its implications.","referenceUrls":["https://jack-clark.net/"],"path":"/entities/jack-clark","json":"/entities/jack-clark.json","markdown":"/entities/jack-clark.md"}]},"researchQuestions":[],"withdrawn":false,"references":{"schemaVersion":1,"target":{"type":"reading","id":"48b9bfa6f6e2078f4e17b78c944aa24a2c67228f29ec80b9f63d329765b64bd9","revision":"aa4be3ed-34a0-abd5-88b5-53e872ad4808","revisionKind":"publication-id","title":"Jack Clark argues in Import AI that AI research taste could bring recursive self-improvement closer","url":"https://escapemeridian.com/readings/jack-clark-argues-in-import-ai-that-ai-research-taste-could-48b9bfa6"},"source":{"type":"source","id":"https://jack-clark.net/2026/10/12/import-ai-476-creativity-and-rsi-worse-than-china-open-weight-models-ai-goes-to-space/","revision":"cb3dd2188128c6892c8012e3c6b03e359737ac196638ff9889895bbc9d7e4a97","revisionKind":"sha256-source-text","title":"Import AI 476: Creativity and RSI; worse-than-China open weight models; AI goes to space","url":"https://jack-clark.net/2026/10/12/import-ai-476-creativity-and-rsi-worse-than-china-open-weight-models-ai-goes-to-space/"},"status":"available","items":[],"next":null,"scope":"Explicit citations in public readings and public file attachments to this reading. Shared topics are not citations. Other artifact types need an explicit reference adapter."},"revisions":[{"id":"48b9bfa6f6e2078f4e17b78c944aa24a2c67228f29ec80b9f63d329765b64bd9","title":"Jack Clark argues in Import AI that AI research taste could bring recursive self-improvement closer","argument":"Import AI was published on October 12, 2026. Jack Clark argues that AI systems are getting better at what researchers call “good scientific taste”: choosing problems, designing experiments and interpreting results. He describes TasteVal, a benchmark from P-Zero Research that tests how models iteratively design experiments and draw conclusions for fixed research problems. Clark reports that its strongest model exceeded the human baseline on the benchmark, at lower cost per attempt. He connects this result to recursive self-improvement, reasoning that creativity may affect how quickly AI systems can develop the capacity to improve themselves. Clark says he now thinks full recursive self-improvement is more likely in 2027 than he had thought before, when he considered it more likely in 2028. This is his forecast, not a reported outcome. TasteVal measures an experimental component of research taste; the account does not establish that benchmark performance will translate into systems improving themselves. The newsletter also discusses visual reasoning benchmarks, space, open-weight models, government computing, and a fictional future in which direct conversations with powerful synthetic minds pose risks to people.","topicIds":["artificial-general-intelligence"],"domain":"other","claimType":"attributed reading","limits":"Text as retrieved on October 23, 2026; originally published October 12, 2026. TasteVal measures an experimental component of research taste for a fixed research problem; the newsletter does not establish that benchmark performance will translate into recursive self-improvement.","speaker":{"name":"Jack Clark","roleAtSource":"Publisher"},"source":{"title":"Import AI 476: Creativity and RSI; worse-than-China open weight models; AI goes to space","publisher":"Jack Clark","format":"Article","url":"https://jack-clark.net/2026/10/12/import-ai-476-creativity-and-rsi-worse-than-china-open-weight-models-ai-goes-to-space/","publishedAt":"2026-10-12T12:30:39.000Z","dateBasis":"Date supplied by the publisher’s feed."},"evidence":[{"text":"Jack Clark reports that TasteVal measures how well a model iteratively designs experiments and draws conclusions from their outcomes for a fixed research problem.","excerpt":"given a fixed research problem, we measure how well a model iteratively designs experiments and draws conclusions from their outcomes","url":"https://jack-clark.net/2026/10/12/import-ai-476-creativity-and-rsi-worse-than-china-open-weight-models-ai-goes-to-space/","relationship":"Source-stated"},{"text":"Clark argues that creativity may be important to how quickly AI systems develop the capacity to improve themselves.","excerpt":"creativity seemed like a key variable for how fast AI systems could develop the capacity to improve themselves.","url":"https://jack-clark.net/2026/10/12/import-ai-476-creativity-and-rsi-worse-than-china-open-weight-models-ai-goes-to-space/","relationship":"Source-stated"},{"text":"Clark says he now considers full recursive self-improvement more likely in 2027 than he did previously; this is his forecast.","excerpt":"I now think it’s more likely we’ll see full RSI become possible in 2027, whereas before I thought it was more likely in 2028.","url":"https://jack-clark.net/2026/10/12/import-ai-476-creativity-and-rsi-worse-than-china-open-weight-models-ai-goes-to-space/","relationship":"Source-stated"}],"sourceMedia":{"status":"unchecked","sourceUrl":"https://jack-clark.net/2026/10/12/import-ai-476-creativity-and-rsi-worse-than-china-open-weight-models-ai-goes-to-space/","checkedAt":null,"items":[]},"assumptions":[],"openQuestion":{"attribution":"Open question","text":"Will improved performance on TasteVal translate into AI systems becoming able to improve themselves?"},"whatWouldChangeView":{"attribution":"Change condition not recorded","text":""},"editorialComparisons":[],"live":{"title":"Jack Clark argues in Import AI that AI research taste could bring recursive self-improvement closer","summary":"Import AI was published on October 12, 2026. Jack Clark argues that AI systems are getting better at what researchers call “good scientific taste”: choosing problems, designing experiments and interpreting results. He describes TasteVal, a benchmark from P-Zero Research that tests how models iteratively design experiments and draw conclusions for fixed research problems. Clark reports that its strongest model exceeded the human baseline on the benchmark, at lower cost per attempt. He connects this result to recursive self-improvement, reasoning that creativity may affect how quickly AI systems can develop the capacity to improve themselves. Clark says he now thinks full recursive self-improvement is more likely in 2027 than he had thought before, when he considered it more likely in 2028. This is his forecast, not a reported outcome. TasteVal measures an experimental component of research taste; the account does not establish that benchmark performance will translate into systems improving themselves. The newsletter also discusses visual reasoning benchmarks, space, open-weight models, government computing, and a fictional future in which direct conversations with powerful synthetic minds pose risks to people.","question":"Will improved performance on TasteVal translate into AI systems becoming able to improve themselves?","claims":[{"text":"Jack Clark reports that TasteVal measures how well a model iteratively designs experiments and draws conclusions from their outcomes for a fixed research problem.","excerpt":"given a fixed research problem, we measure how well a model iteratively designs experiments and draws conclusions from their outcomes","support":"source-stated"},{"text":"Clark argues that creativity may be important to how quickly AI systems develop the capacity to improve themselves.","excerpt":"creativity seemed like a key variable for how fast AI systems could develop the capacity to improve themselves.","support":"source-stated"},{"text":"Clark says he now considers full recursive self-improvement more likely in 2027 than he did previously; this is his forecast.","excerpt":"I now think it’s more likely we’ll see full RSI become possible in 2027, whereas before I thought it was more likely in 2028.","support":"source-stated"}],"limitations":["Text as retrieved on October 23, 2026; originally published October 12, 2026.","TasteVal measures an experimental component of research taste for a fixed research problem; the newsletter does not establish that benchmark performance will translate into recursive self-improvement."],"mediaId":null,"correctionReason":null,"id":"48b9bfa6f6e2078f4e17b78c944aa24a2c67228f29ec80b9f63d329765b64bd9","publicationId":"aa4be3ed-34a0-abd5-88b5-53e872ad4808","version":1,"sourceUrl":"https://jack-clark.net/2026/10/12/import-ai-476-creativity-and-rsi-worse-than-china-open-weight-models-ai-goes-to-space/","sourceTitle":"Import AI 476: Creativity and RSI; worse-than-China open weight models; AI goes to space","publisher":"Jack Clark","publishedAt":"2026-10-12T12:30:39.000Z","originalPublished":{"value":"2026-10-12T12:30:39.000Z","precision":"instant","basis":"backfill:feed_published"},"firstObservedAt":"2026-10-23T20:01:48.876Z","createdAt":"2026-10-31T11:34:28.474Z","historical":true,"entityIds":["jack-clark"],"topicIds":["artificial-general-intelligence"],"media":[],"sourceMedia":{"status":"unchecked","sourceUrl":"https://jack-clark.net/2026/10/12/import-ai-476-creativity-and-rsi-worse-than-china-open-weight-models-ai-goes-to-space/","checkedAt":null,"items":[]},"editorialDecision":{"outcome":"accept","mediaId":null,"inputHash":"14af1025635d34a870dbd0ba3c157d4c96b30569df438b70d12eeef756ac479c","rubricVersion":"unrecorded","provider":"unrecorded","model":"anthropic/claude-opus-5.5","reasonCodes":["bridge_admitted_by_model","second_opinion_categorical"],"confidence":1,"decidedAt":"2026-10-23T20:01:46.020Z"},"provenance":{"textHash":"cb3dd2188128c6892c8012e3c6b03e359737ac196638ff9889895bbc9d7e4a97","observationId":"7d1dafa435e87dc1b5bba98375af4282c718eea23d5be2bc60ed499c9973ddb3","retrievedAt":"2026-10-23T20:03:37.266Z","kind":"publisher-page","preparer":{"model":"openai/gpt-6-luna","runId":"425a5e3f-9304-029e-dbbc-71514e0257ab"},"reviewer":{"model":"anthropic/claude-opus-5.5","runId":"39c956c8-6df2-5248-b9de-9713bc79377a"}},"audit":{"status":"passed","reviewedAt":"2026-10-31T11:34:28.165Z","model":"anthropic/claude-opus-5.5"}}}],"audits":[{"publicationId":"aa4be3ed-34a0-abd5-88b5-53e872ad4808","verdict":"passed","findings":[{"dimension":"attribution","status":"pass","explanation":"The reading attributes the newsletter and its forecast to Jack Clark. It attributes TasteVal to P-Zero Research and presents the benchmark results as reported, not as fact. The source is one newsletter issue with the author's own commentary, not a bare listing.","excerpt":"according to a new benchmark from P-Zero Research called TasteVal"},{"dimension":"summary","status":"pass","explanation":"The reading leads with Clark's argument and explains research taste in plain terms. It describes what TasteVal measures and reports Clark's revised RSI timeline accurately. Recursive self-improvement and benchmark are defined beside the reading. The question is substantive and open. The closing note on the other newsletter sections is accurate and labels the final story as fiction.","excerpt":"I now think it’s more likely we’ll see full RSI become possible in 2027, whereas before I thought it was more likely in 2028."},{"dimension":"date","status":"pass","explanation":"The reading states October 12, 2026, which matches the UTC calendar day of evidence.originalPublished (2026-10-12T12:30:39Z). The 2027 and 2028 RSI dates are presented as Clark's forecast, not outcomes. The 2029 setting appears only in fiction, which the reading labels as fictional. The evidence is a publisher page, so no archive disclosure is needed.","excerpt":null},{"dimension":"support","status":"pass","explanation":"Each claim matches the source text. The source says the best model significantly exceeds the human baseline at roughly 1/30 of the experts' cost per attempt, which supports the lower-cost claim. The limitation about translating benchmark results into self-improvement is suitably cautious. No unsupported benefits or costs are implied.","excerpt":"The best model, Opus 5.5, significantly exceeds our human baseline with a compute multiplier of 2.30x (95% CI 1.15-4.37x), at roughly 1/30 of our expert baseliners’ average cost per attempt."},{"dimension":"media","status":"pass","explanation":"No media is selected, and no verified media candidates were supplied, so nothing is misidentified or invented.","excerpt":null}],"createdAt":"2026-10-31T11:34:28.165Z","model":"anthropic/claude-opus-5.5"}],"correctionUrl":"https://escapemeridian.com/corrections/"}