{"id":"d36a99482970d9b99d1468cf8866fd76591d4b4f1c98b479fecffd4f59c670a8","title":"arXiv cs.AI: Re-Search and how LLMs seek research ideas","argument":"In the Re-Search preprint, Muhan Gao and coauthors study how LLMs develop research ideas through a two-player questioning game. Given the research problem of a published paper, a model tries to rediscover its core idea by asking yes-or-no questions and receiving binary oracle feedback. The authors say this setup tracks how ideas change while isolating ideation from execution.\n\nBuilt from 200 ICML 2026 papers, the evaluation covers six recent LLMs. The authors report that nearly all progress happens within the first 30 of 100 questions, after which most models plateau. Their controlled interventions suggest a gap between recognizing and generating: models can identify a good idea or question when offered one, but often cannot produce it themselves. The authors also find two patterns of failure: pursuing one direction too heavily or exploring too many.\n\nThe authors argue these results point toward enriching autoresearch systems with externally sourced ideas and questions, rather than relying only on improvements to a model’s generation ability. This is a proposed design principle from the preprint, not a demonstrated improvement in research outcomes.","topicIds":["intelligence-and-discovery"],"domain":"science","claimType":"attributed reading","limits":"The evidence provided is the abstract of a preprint; it does not include the paper’s detailed methods or results. The authors’ proposed design principle is framed as a possibility; the abstract does not establish its effect in practical autoresearch systems. ","speaker":{"name":"Gao, Muhan, Zhu, Hangxiao, Su, Xinqi, Zhang, Yu, Khashabi, Daniel, Huang, Kuan-Hao","roleAtSource":"Article author"},"source":{"title":"Re-Search: Dissecting LLM Ideation Through a Two-Player Questioning Game","publisher":"arXiv cs.AI","format":"Article","url":"https://arxiv.org/abs/2611.14487","publishedAt":"2026-11-10","dateBasis":"Date basis not provided."},"evidence":[{"text":"The authors introduce Re-Search, a questioning-game evaluation in which a model tries to rediscover a published paper’s core idea from its research problem.","excerpt":"given only the research problem of a published paper, a model must rediscover the paper's core idea by asking yes-or-no questions","url":"https://arxiv.org/abs/2611.14487","relationship":"Source-stated"},{"text":"The authors say the evaluation tracks changing ideas while isolating ideation from execution.","excerpt":"Re-Search tracks how a model's idea evolves as it asks questions and receives binary oracle feedback, isolating ideation from execution.","url":"https://arxiv.org/abs/2611.14487","relationship":"Source-stated"},{"text":"In the evaluation built from 200 ICML 2026 papers, the authors report that nearly all progress by six recent LLMs happens within the first 30 of 100 questions; most models then plateau.","excerpt":"Across six recent LLMs, nearly all progress happens within the first 30 of 100 questions, which largely decide how the game ends; after that, most models plateau.","url":"https://arxiv.org/abs/2611.14487","relationship":"Source-stated"},{"text":"The authors attribute the bottleneck to generating ideas and questions, rather than judging candidates when they are provided.","excerpt":"the bottleneck is generation, not judgment: models recognize a good idea or a good question when one is offered, but often cannot produce it on their own.","url":"https://arxiv.org/abs/2611.14487","relationship":"Source-stated"},{"text":"The authors suggest enriching autoresearch systems with externally sourced ideas and questions; the abstract does not report that this approach improves research outcomes in practice.","excerpt":"enriching the search with externally sourced ideas and questions may be more effective than improving any single model's generation ability.","url":"https://arxiv.org/abs/2611.14487","relationship":"Source-stated"}],"sourceMedia":{"status":"not_found","sourceUrl":"https://arxiv.org/abs/2611.14487","checkedAt":"2026-11-12T03:20:09.956Z","items":[],"note":"No suitable media was found by the supported page extraction."},"assumptions":[],"openQuestion":{"attribution":"Open question","text":"How would externally sourced ideas and questions affect the search in Re-Search, given the authors’ finding that models can recognize candidates they often cannot generate?"},"whatWouldChangeView":{"attribution":"Change condition not recorded","text":""},"editorialComparisons":[],"live":{"title":"arXiv cs.AI: Re-Search and how LLMs seek research ideas","summary":"In the Re-Search preprint, Muhan Gao and coauthors study how LLMs develop research ideas through a two-player questioning game. Given the research problem of a published paper, a model tries to rediscover its core idea by asking yes-or-no questions and receiving binary oracle feedback. The authors say this setup tracks how ideas change while isolating ideation from execution.\n\nBuilt from 200 ICML 2026 papers, the evaluation covers six recent LLMs. The authors report that nearly all progress happens within the first 30 of 100 questions, after which most models plateau. Their controlled interventions suggest a gap between recognizing and generating: models can identify a good idea or question when offered one, but often cannot produce it themselves. The authors also find two patterns of failure: pursuing one direction too heavily or exploring too many.\n\nThe authors argue these results point toward enriching autoresearch systems with externally sourced ideas and questions, rather than relying only on improvements to a model’s generation ability. This is a proposed design principle from the preprint, not a demonstrated improvement in research outcomes.","question":"How would externally sourced ideas and questions affect the search in Re-Search, given the authors’ finding that models can recognize candidates they often cannot generate?","claims":[{"text":"The authors introduce Re-Search, a questioning-game evaluation in which a model tries to rediscover a published paper’s core idea from its research problem.","excerpt":"given only the research problem of a published paper, a model must rediscover the paper's core idea by asking yes-or-no questions","support":"source-stated"},{"text":"The authors say the evaluation tracks changing ideas while isolating ideation from execution.","excerpt":"Re-Search tracks how a model's idea evolves as it asks questions and receives binary oracle feedback, isolating ideation from execution.","support":"source-stated"},{"text":"In the evaluation built from 200 ICML 2026 papers, the authors report that nearly all progress by six recent LLMs happens within the first 30 of 100 questions; most models then plateau.","excerpt":"Across six recent LLMs, nearly all progress happens within the first 30 of 100 questions, which largely decide how the game ends; after that, most models plateau.","support":"source-stated"},{"text":"The authors attribute the bottleneck to generating ideas and questions, rather than judging candidates when they are provided.","excerpt":"the bottleneck is generation, not judgment: models recognize a good idea or a good question when one is offered, but often cannot produce it on their own.","support":"source-stated"},{"text":"The authors suggest enriching autoresearch systems with externally sourced ideas and questions; the abstract does not report that this approach improves research outcomes in practice.","excerpt":"enriching the search with externally sourced ideas and questions may be more effective than improving any single model's generation ability.","support":"source-stated"}],"limitations":["The evidence provided is the abstract of a preprint; it does not include the paper’s detailed methods or results.","The authors’ proposed design principle is framed as a possibility; the abstract does not establish its effect in practical autoresearch systems. "],"mediaId":"4276c0cc3fd5f0877abaee3f8cfcd449e8af7be0a9b7743c46c7dee4a3f2de22","correctionReason":null,"id":"d36a99482970d9b99d1468cf8866fd76591d4b4f1c98b479fecffd4f59c670a8","publicationId":"61c7d1a4-65a6-4e56-8bfb-1021b18c7ef0","version":1,"sourceUrl":"https://arxiv.org/abs/2611.14487","sourceTitle":"Re-Search: Dissecting LLM Ideation Through a Two-Player Questioning Game","publisher":"arXiv cs.AI","publishedAt":null,"originalPublished":{"value":"2026-11-10","precision":"day","basis":"publisher_citation_date"},"firstObservedAt":"2026-11-11T05:04:40.474Z","createdAt":"2026-11-12T02:17:11.268Z","historical":false,"entityIds":["arxiv"],"topicIds":["intelligence-and-discovery"],"media":[{"id":"4276c0cc3fd5f0877abaee3f8cfcd449e8af7be0a9b7743c46c7dee4a3f2de22","kind":"article","url":"https://arxiv.org/abs/2611.14487","sourceUrl":"https://arxiv.org/abs/2611.14487","label":"Read original","provenance":"source_link","verification":{"status":"verified","checkedAt":"2026-11-12T02:03:17.007Z","method":"acquired_source","contentType":"text/html"},"recommendationReason":"Recommended by Jev for understanding this source; checked by Luna."}],"sourceMedia":{"status":"not_found","sourceUrl":"https://arxiv.org/abs/2611.14487","checkedAt":"2026-11-12T03:20:09.956Z","items":[],"note":"No suitable media was found by the supported page extraction."},"editorialDecision":{"outcome":"accept","reasonCodes":["material_development"],"rubricVersion":"stream-admission-v6","mediaId":"4276c0cc3fd5f0877abaee3f8cfcd449e8af7be0a9b7743c46c7dee4a3f2de22","confidence":0.83,"provider":"typesafe","model":"jev-1.13.0","inputHash":"bbc2ed2524653f45ed6abe2f85d18f51638b1a3324408d8ead1d18c2d41882c2","decidedAt":"2026-11-12T02:03:18.538Z"},"provenance":{"textHash":"68e63692816212872d095cc392ab85f3f45173152d4a7dea94c8f52bad67a57e","observationId":"27f69a60def8ebb68577826b6c9f51263012e0127f66db4c958e5fb9f4e32636","retrievedAt":"2026-11-12T02:03:16.820Z","kind":"publisher-page","preparer":{"model":"gpt-6-luna","runId":"5da2b822-93f2-4e11-8c09-54a01f05789b"},"reviewer":{"model":"gpt-6-luna","runId":"f1f245f3-ad20-46fd-87c0-003cb1e56689"}},"audit":{"status":"passed","reviewedAt":"2026-11-12T02:17:11.268Z","model":"anthropic/claude-opus-5.5"},"sourceAttribution":{"authors":[{"name":"Gao, Muhan","url":null,"kind":"unknown"},{"name":"Zhu, Hangxiao","url":null,"kind":"unknown"},{"name":"Su, Xinqi","url":null,"kind":"unknown"},{"name":"Zhang, Yu","url":null,"kind":"unknown"},{"name":"Khashabi, Daniel","url":null,"kind":"unknown"},{"name":"Huang, Kuan-Hao","url":null,"kind":"unknown"}],"authorRole":"article-author","hosts":[],"status":"stated","authorRoster":{"truncated":false,"omitted":0,"assertionGroupsOmitted":0},"participation":[{"actor":{"name":"Gao, Muhan","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Gao, Muhan","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2611.14487","locator":"head/meta[name=author|citation_author]"},"assertionId":"dbfe70017df89d7f86b427777adf0a9ad54e575a770fa5e318c48d62de2001c0:participation:0","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-12T03:20:09.956Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"dbfe70017df89d7f86b427777adf0a9ad54e575a770fa5e318c48d62de2001c0","sourceUrl":"https://arxiv.org/abs/2611.14487","contentSha256":"bc3b6649ad0074c58ed167a54e0d0f676d786d4af77d990ad6c494e02d5cc3e1","textHash":"68e63692816212872d095cc392ab85f3f45173152d4a7dea94c8f52bad67a57e","observedAt":"2026-11-12T03:20:09.956Z"}},{"actor":{"name":"Zhu, Hangxiao","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Zhu, Hangxiao","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2611.14487","locator":"head/meta[name=author|citation_author]"},"assertionId":"dbfe70017df89d7f86b427777adf0a9ad54e575a770fa5e318c48d62de2001c0:participation:1","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-12T03:20:09.956Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"dbfe70017df89d7f86b427777adf0a9ad54e575a770fa5e318c48d62de2001c0","sourceUrl":"https://arxiv.org/abs/2611.14487","contentSha256":"bc3b6649ad0074c58ed167a54e0d0f676d786d4af77d990ad6c494e02d5cc3e1","textHash":"68e63692816212872d095cc392ab85f3f45173152d4a7dea94c8f52bad67a57e","observedAt":"2026-11-12T03:20:09.956Z"}},{"actor":{"name":"Su, Xinqi","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Su, Xinqi","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2611.14487","locator":"head/meta[name=author|citation_author]"},"assertionId":"dbfe70017df89d7f86b427777adf0a9ad54e575a770fa5e318c48d62de2001c0:participation:2","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-12T03:20:09.956Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"dbfe70017df89d7f86b427777adf0a9ad54e575a770fa5e318c48d62de2001c0","sourceUrl":"https://arxiv.org/abs/2611.14487","contentSha256":"bc3b6649ad0074c58ed167a54e0d0f676d786d4af77d990ad6c494e02d5cc3e1","textHash":"68e63692816212872d095cc392ab85f3f45173152d4a7dea94c8f52bad67a57e","observedAt":"2026-11-12T03:20:09.956Z"}},{"actor":{"name":"Zhang, Yu","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Zhang, Yu","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2611.14487","locator":"head/meta[name=author|citation_author]"},"assertionId":"dbfe70017df89d7f86b427777adf0a9ad54e575a770fa5e318c48d62de2001c0:participation:3","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-12T03:20:09.956Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"dbfe70017df89d7f86b427777adf0a9ad54e575a770fa5e318c48d62de2001c0","sourceUrl":"https://arxiv.org/abs/2611.14487","contentSha256":"bc3b6649ad0074c58ed167a54e0d0f676d786d4af77d990ad6c494e02d5cc3e1","textHash":"68e63692816212872d095cc392ab85f3f45173152d4a7dea94c8f52bad67a57e","observedAt":"2026-11-12T03:20:09.956Z"}},{"actor":{"name":"Khashabi, Daniel","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Khashabi, Daniel","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2611.14487","locator":"head/meta[name=author|citation_author]"},"assertionId":"dbfe70017df89d7f86b427777adf0a9ad54e575a770fa5e318c48d62de2001c0:participation:4","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-12T03:20:09.956Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"dbfe70017df89d7f86b427777adf0a9ad54e575a770fa5e318c48d62de2001c0","sourceUrl":"https://arxiv.org/abs/2611.14487","contentSha256":"bc3b6649ad0074c58ed167a54e0d0f676d786d4af77d990ad6c494e02d5cc3e1","textHash":"68e63692816212872d095cc392ab85f3f45173152d4a7dea94c8f52bad67a57e","observedAt":"2026-11-12T03:20:09.956Z"}},{"actor":{"name":"Huang, Kuan-Hao","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Huang, Kuan-Hao","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2611.14487","locator":"head/meta[name=author|citation_author]"},"assertionId":"dbfe70017df89d7f86b427777adf0a9ad54e575a770fa5e318c48d62de2001c0:participation:5","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-12T03:20:09.956Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"dbfe70017df89d7f86b427777adf0a9ad54e575a770fa5e318c48d62de2001c0","sourceUrl":"https://arxiv.org/abs/2611.14487","contentSha256":"bc3b6649ad0074c58ed167a54e0d0f676d786d4af77d990ad6c494e02d5cc3e1","textHash":"68e63692816212872d095cc392ab85f3f45173152d4a7dea94c8f52bad67a57e","observedAt":"2026-11-12T03:20:09.956Z"}}],"participationCoverage":{"total":6,"returned":6,"complete":true,"fullAssertions":"source-metadata-endpoint"},"distribution":null,"publisher":null,"sourceUrl":"https://arxiv.org/abs/2611.14487","snapshotId":"dbfe70017df89d7f86b427777adf0a9ad54e575a770fa5e318c48d62de2001c0","contentSha256":"bc3b6649ad0074c58ed167a54e0d0f676d786d4af77d990ad6c494e02d5cc3e1","textHash":"68e63692816212872d095cc392ab85f3f45173152d4a7dea94c8f52bad67a57e","observedAt":"2026-11-12T03:20:09.956Z","basis":"publisher-page-metadata","sourceLinks":44,"linksTruncated":false,"metadataUrl":"/api/pipeline/stream/d36a99482970d9b99d1468cf8866fd76591d4b4f1c98b479fecffd4f59c670a8/source"}},"canonicalUrl":"https://escapemeridian.com/readings/arxiv-cs-ai-re-search-and-how-llms-seek-research-ideas-d36a9948","formats":{"html":"/readings/arxiv-cs-ai-re-search-and-how-llms-seek-research-ideas-d36a9948","markdown":"/readings/arxiv-cs-ai-re-search-and-how-llms-seek-research-ideas-d36a9948.md","json":"/readings/arxiv-cs-ai-re-search-and-how-llms-seek-research-ideas-d36a9948.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.","maintainedGuideLinks":[],"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-llm","name":"Large language model","explanation":"A language model with many learned numerical settings, used to process and generate language.","referenceUrl":"https://developers.google.com/machine-learning/glossary#large-language-model"}],"entities":[{"id":"arxiv","name":"arXiv","kind":"organization","context":"arXiv is a research manuscript repository where authors share papers that may not yet have been peer reviewed.","referenceUrls":["https://arxiv.org/","https://info.arxiv.org/help/api/index.html"],"path":"/entities/arxiv","json":"/entities/arxiv.json","markdown":"/entities/arxiv.md"}]},"researchQuestions":[],"withdrawn":false,"references":{"schemaVersion":1,"target":{"type":"reading","id":"d36a99482970d9b99d1468cf8866fd76591d4b4f1c98b479fecffd4f59c670a8","revision":"61c7d1a4-65a6-4e56-8bfb-1021b18c7ef0","revisionKind":"publication-id","title":"arXiv cs.AI: Re-Search and how LLMs seek research ideas","url":"https://escapemeridian.com/readings/arxiv-cs-ai-re-search-and-how-llms-seek-research-ideas-d36a9948"},"source":{"type":"source","id":"https://arxiv.org/abs/2611.14487","revision":"68e63692816212872d095cc392ab85f3f45173152d4a7dea94c8f52bad67a57e","revisionKind":"sha256-source-text","title":"Re-Search: Dissecting LLM Ideation Through a Two-Player Questioning Game","url":"https://arxiv.org/abs/2611.14487"},"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":"d36a99482970d9b99d1468cf8866fd76591d4b4f1c98b479fecffd4f59c670a8","title":"arXiv cs.AI: Re-Search and how LLMs seek research ideas","argument":"In the Re-Search preprint, Muhan Gao and coauthors study how LLMs develop research ideas through a two-player questioning game. Given the research problem of a published paper, a model tries to rediscover its core idea by asking yes-or-no questions and receiving binary oracle feedback. The authors say this setup tracks how ideas change while isolating ideation from execution.\n\nBuilt from 200 ICML 2026 papers, the evaluation covers six recent LLMs. The authors report that nearly all progress happens within the first 30 of 100 questions, after which most models plateau. Their controlled interventions suggest a gap between recognizing and generating: models can identify a good idea or question when offered one, but often cannot produce it themselves. The authors also find two patterns of failure: pursuing one direction too heavily or exploring too many.\n\nThe authors argue these results point toward enriching autoresearch systems with externally sourced ideas and questions, rather than relying only on improvements to a model’s generation ability. This is a proposed design principle from the preprint, not a demonstrated improvement in research outcomes.","topicIds":["intelligence-and-discovery"],"domain":"science","claimType":"attributed reading","limits":"The evidence provided is the abstract of a preprint; it does not include the paper’s detailed methods or results. The authors’ proposed design principle is framed as a possibility; the abstract does not establish its effect in practical autoresearch systems. ","speaker":{"name":"Gao, Muhan, Zhu, Hangxiao, Su, Xinqi, Zhang, Yu, Khashabi, Daniel, Huang, Kuan-Hao","roleAtSource":"Article author"},"source":{"title":"Re-Search: Dissecting LLM Ideation Through a Two-Player Questioning Game","publisher":"arXiv cs.AI","format":"Article","url":"https://arxiv.org/abs/2611.14487","publishedAt":"2026-11-10","dateBasis":"Date basis not provided."},"evidence":[{"text":"The authors introduce Re-Search, a questioning-game evaluation in which a model tries to rediscover a published paper’s core idea from its research problem.","excerpt":"given only the research problem of a published paper, a model must rediscover the paper's core idea by asking yes-or-no questions","url":"https://arxiv.org/abs/2611.14487","relationship":"Source-stated"},{"text":"The authors say the evaluation tracks changing ideas while isolating ideation from execution.","excerpt":"Re-Search tracks how a model's idea evolves as it asks questions and receives binary oracle feedback, isolating ideation from execution.","url":"https://arxiv.org/abs/2611.14487","relationship":"Source-stated"},{"text":"In the evaluation built from 200 ICML 2026 papers, the authors report that nearly all progress by six recent LLMs happens within the first 30 of 100 questions; most models then plateau.","excerpt":"Across six recent LLMs, nearly all progress happens within the first 30 of 100 questions, which largely decide how the game ends; after that, most models plateau.","url":"https://arxiv.org/abs/2611.14487","relationship":"Source-stated"},{"text":"The authors attribute the bottleneck to generating ideas and questions, rather than judging candidates when they are provided.","excerpt":"the bottleneck is generation, not judgment: models recognize a good idea or a good question when one is offered, but often cannot produce it on their own.","url":"https://arxiv.org/abs/2611.14487","relationship":"Source-stated"},{"text":"The authors suggest enriching autoresearch systems with externally sourced ideas and questions; the abstract does not report that this approach improves research outcomes in practice.","excerpt":"enriching the search with externally sourced ideas and questions may be more effective than improving any single model's generation ability.","url":"https://arxiv.org/abs/2611.14487","relationship":"Source-stated"}],"sourceMedia":{"status":"not_found","sourceUrl":"https://arxiv.org/abs/2611.14487","checkedAt":"2026-11-12T03:20:09.956Z","items":[],"note":"No suitable media was found by the supported page extraction."},"assumptions":[],"openQuestion":{"attribution":"Open question","text":"How would externally sourced ideas and questions affect the search in Re-Search, given the authors’ finding that models can recognize candidates they often cannot generate?"},"whatWouldChangeView":{"attribution":"Change condition not recorded","text":""},"editorialComparisons":[],"live":{"title":"arXiv cs.AI: Re-Search and how LLMs seek research ideas","summary":"In the Re-Search preprint, Muhan Gao and coauthors study how LLMs develop research ideas through a two-player questioning game. Given the research problem of a published paper, a model tries to rediscover its core idea by asking yes-or-no questions and receiving binary oracle feedback. The authors say this setup tracks how ideas change while isolating ideation from execution.\n\nBuilt from 200 ICML 2026 papers, the evaluation covers six recent LLMs. The authors report that nearly all progress happens within the first 30 of 100 questions, after which most models plateau. Their controlled interventions suggest a gap between recognizing and generating: models can identify a good idea or question when offered one, but often cannot produce it themselves. The authors also find two patterns of failure: pursuing one direction too heavily or exploring too many.\n\nThe authors argue these results point toward enriching autoresearch systems with externally sourced ideas and questions, rather than relying only on improvements to a model’s generation ability. This is a proposed design principle from the preprint, not a demonstrated improvement in research outcomes.","question":"How would externally sourced ideas and questions affect the search in Re-Search, given the authors’ finding that models can recognize candidates they often cannot generate?","claims":[{"text":"The authors introduce Re-Search, a questioning-game evaluation in which a model tries to rediscover a published paper’s core idea from its research problem.","excerpt":"given only the research problem of a published paper, a model must rediscover the paper's core idea by asking yes-or-no questions","support":"source-stated"},{"text":"The authors say the evaluation tracks changing ideas while isolating ideation from execution.","excerpt":"Re-Search tracks how a model's idea evolves as it asks questions and receives binary oracle feedback, isolating ideation from execution.","support":"source-stated"},{"text":"In the evaluation built from 200 ICML 2026 papers, the authors report that nearly all progress by six recent LLMs happens within the first 30 of 100 questions; most models then plateau.","excerpt":"Across six recent LLMs, nearly all progress happens within the first 30 of 100 questions, which largely decide how the game ends; after that, most models plateau.","support":"source-stated"},{"text":"The authors attribute the bottleneck to generating ideas and questions, rather than judging candidates when they are provided.","excerpt":"the bottleneck is generation, not judgment: models recognize a good idea or a good question when one is offered, but often cannot produce it on their own.","support":"source-stated"},{"text":"The authors suggest enriching autoresearch systems with externally sourced ideas and questions; the abstract does not report that this approach improves research outcomes in practice.","excerpt":"enriching the search with externally sourced ideas and questions may be more effective than improving any single model's generation ability.","support":"source-stated"}],"limitations":["The evidence provided is the abstract of a preprint; it does not include the paper’s detailed methods or results.","The authors’ proposed design principle is framed as a possibility; the abstract does not establish its effect in practical autoresearch systems. "],"mediaId":"4276c0cc3fd5f0877abaee3f8cfcd449e8af7be0a9b7743c46c7dee4a3f2de22","correctionReason":null,"id":"d36a99482970d9b99d1468cf8866fd76591d4b4f1c98b479fecffd4f59c670a8","publicationId":"61c7d1a4-65a6-4e56-8bfb-1021b18c7ef0","version":1,"sourceUrl":"https://arxiv.org/abs/2611.14487","sourceTitle":"Re-Search: Dissecting LLM Ideation Through a Two-Player Questioning Game","publisher":"arXiv cs.AI","publishedAt":null,"originalPublished":{"value":"2026-11-10","precision":"day","basis":"publisher_citation_date"},"firstObservedAt":"2026-11-11T05:04:40.474Z","createdAt":"2026-11-12T02:17:11.268Z","historical":false,"entityIds":["arxiv"],"topicIds":["intelligence-and-discovery"],"media":[{"id":"4276c0cc3fd5f0877abaee3f8cfcd449e8af7be0a9b7743c46c7dee4a3f2de22","kind":"article","url":"https://arxiv.org/abs/2611.14487","sourceUrl":"https://arxiv.org/abs/2611.14487","label":"Read original","provenance":"source_link","verification":{"status":"verified","checkedAt":"2026-11-12T02:03:17.007Z","method":"acquired_source","contentType":"text/html"},"recommendationReason":"Recommended by Jev for understanding this source; checked by Luna."}],"sourceMedia":{"status":"not_found","sourceUrl":"https://arxiv.org/abs/2611.14487","checkedAt":"2026-11-12T03:20:09.956Z","items":[],"note":"No suitable media was found by the supported page extraction."},"editorialDecision":{"outcome":"accept","reasonCodes":["material_development"],"rubricVersion":"stream-admission-v6","mediaId":"4276c0cc3fd5f0877abaee3f8cfcd449e8af7be0a9b7743c46c7dee4a3f2de22","confidence":0.83,"provider":"typesafe","model":"jev-1.13.0","inputHash":"bbc2ed2524653f45ed6abe2f85d18f51638b1a3324408d8ead1d18c2d41882c2","decidedAt":"2026-11-12T02:03:18.538Z"},"provenance":{"textHash":"68e63692816212872d095cc392ab85f3f45173152d4a7dea94c8f52bad67a57e","observationId":"27f69a60def8ebb68577826b6c9f51263012e0127f66db4c958e5fb9f4e32636","retrievedAt":"2026-11-12T02:03:16.820Z","kind":"publisher-page","preparer":{"model":"gpt-6-luna","runId":"5da2b822-93f2-4e11-8c09-54a01f05789b"},"reviewer":{"model":"gpt-6-luna","runId":"f1f245f3-ad20-46fd-87c0-003cb1e56689"}},"audit":{"status":"passed","reviewedAt":"2026-11-12T02:17:11.268Z","model":"anthropic/claude-opus-5.5"},"sourceAttribution":{"authors":[{"name":"Gao, Muhan","url":null,"kind":"unknown"},{"name":"Zhu, Hangxiao","url":null,"kind":"unknown"},{"name":"Su, Xinqi","url":null,"kind":"unknown"},{"name":"Zhang, Yu","url":null,"kind":"unknown"},{"name":"Khashabi, Daniel","url":null,"kind":"unknown"},{"name":"Huang, Kuan-Hao","url":null,"kind":"unknown"}],"authorRole":"article-author","hosts":[],"status":"stated","authorRoster":{"truncated":false,"omitted":0,"assertionGroupsOmitted":0},"participation":[{"actor":{"name":"Gao, Muhan","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Gao, Muhan","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2611.14487","locator":"head/meta[name=author|citation_author]"},"assertionId":"dbfe70017df89d7f86b427777adf0a9ad54e575a770fa5e318c48d62de2001c0:participation:0","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-12T03:20:09.956Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"dbfe70017df89d7f86b427777adf0a9ad54e575a770fa5e318c48d62de2001c0","sourceUrl":"https://arxiv.org/abs/2611.14487","contentSha256":"bc3b6649ad0074c58ed167a54e0d0f676d786d4af77d990ad6c494e02d5cc3e1","textHash":"68e63692816212872d095cc392ab85f3f45173152d4a7dea94c8f52bad67a57e","observedAt":"2026-11-12T03:20:09.956Z"}},{"actor":{"name":"Zhu, Hangxiao","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Zhu, Hangxiao","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2611.14487","locator":"head/meta[name=author|citation_author]"},"assertionId":"dbfe70017df89d7f86b427777adf0a9ad54e575a770fa5e318c48d62de2001c0:participation:1","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-12T03:20:09.956Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"dbfe70017df89d7f86b427777adf0a9ad54e575a770fa5e318c48d62de2001c0","sourceUrl":"https://arxiv.org/abs/2611.14487","contentSha256":"bc3b6649ad0074c58ed167a54e0d0f676d786d4af77d990ad6c494e02d5cc3e1","textHash":"68e63692816212872d095cc392ab85f3f45173152d4a7dea94c8f52bad67a57e","observedAt":"2026-11-12T03:20:09.956Z"}},{"actor":{"name":"Su, Xinqi","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Su, Xinqi","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2611.14487","locator":"head/meta[name=author|citation_author]"},"assertionId":"dbfe70017df89d7f86b427777adf0a9ad54e575a770fa5e318c48d62de2001c0:participation:2","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-12T03:20:09.956Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"dbfe70017df89d7f86b427777adf0a9ad54e575a770fa5e318c48d62de2001c0","sourceUrl":"https://arxiv.org/abs/2611.14487","contentSha256":"bc3b6649ad0074c58ed167a54e0d0f676d786d4af77d990ad6c494e02d5cc3e1","textHash":"68e63692816212872d095cc392ab85f3f45173152d4a7dea94c8f52bad67a57e","observedAt":"2026-11-12T03:20:09.956Z"}},{"actor":{"name":"Zhang, Yu","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Zhang, Yu","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2611.14487","locator":"head/meta[name=author|citation_author]"},"assertionId":"dbfe70017df89d7f86b427777adf0a9ad54e575a770fa5e318c48d62de2001c0:participation:3","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-12T03:20:09.956Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"dbfe70017df89d7f86b427777adf0a9ad54e575a770fa5e318c48d62de2001c0","sourceUrl":"https://arxiv.org/abs/2611.14487","contentSha256":"bc3b6649ad0074c58ed167a54e0d0f676d786d4af77d990ad6c494e02d5cc3e1","textHash":"68e63692816212872d095cc392ab85f3f45173152d4a7dea94c8f52bad67a57e","observedAt":"2026-11-12T03:20:09.956Z"}},{"actor":{"name":"Khashabi, Daniel","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Khashabi, Daniel","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2611.14487","locator":"head/meta[name=author|citation_author]"},"assertionId":"dbfe70017df89d7f86b427777adf0a9ad54e575a770fa5e318c48d62de2001c0:participation:4","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-12T03:20:09.956Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"dbfe70017df89d7f86b427777adf0a9ad54e575a770fa5e318c48d62de2001c0","sourceUrl":"https://arxiv.org/abs/2611.14487","contentSha256":"bc3b6649ad0074c58ed167a54e0d0f676d786d4af77d990ad6c494e02d5cc3e1","textHash":"68e63692816212872d095cc392ab85f3f45173152d4a7dea94c8f52bad67a57e","observedAt":"2026-11-12T03:20:09.956Z"}},{"actor":{"name":"Huang, Kuan-Hao","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Huang, Kuan-Hao","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2611.14487","locator":"head/meta[name=author|citation_author]"},"assertionId":"dbfe70017df89d7f86b427777adf0a9ad54e575a770fa5e318c48d62de2001c0:participation:5","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-12T03:20:09.956Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"dbfe70017df89d7f86b427777adf0a9ad54e575a770fa5e318c48d62de2001c0","sourceUrl":"https://arxiv.org/abs/2611.14487","contentSha256":"bc3b6649ad0074c58ed167a54e0d0f676d786d4af77d990ad6c494e02d5cc3e1","textHash":"68e63692816212872d095cc392ab85f3f45173152d4a7dea94c8f52bad67a57e","observedAt":"2026-11-12T03:20:09.956Z"}}],"participationCoverage":{"total":6,"returned":6,"complete":true,"fullAssertions":"source-metadata-endpoint"},"distribution":null,"publisher":null,"sourceUrl":"https://arxiv.org/abs/2611.14487","snapshotId":"dbfe70017df89d7f86b427777adf0a9ad54e575a770fa5e318c48d62de2001c0","contentSha256":"bc3b6649ad0074c58ed167a54e0d0f676d786d4af77d990ad6c494e02d5cc3e1","textHash":"68e63692816212872d095cc392ab85f3f45173152d4a7dea94c8f52bad67a57e","observedAt":"2026-11-12T03:20:09.956Z","basis":"publisher-page-metadata","sourceLinks":44,"linksTruncated":false,"metadataUrl":"/api/pipeline/stream/d36a99482970d9b99d1468cf8866fd76591d4b4f1c98b479fecffd4f59c670a8/source"}}}],"audits":[{"publicationId":"61c7d1a4-65a6-4e56-8bfb-1021b18c7ef0","verdict":"passed","findings":[{"dimension":"attribution","status":"pass","explanation":"The reading attributes the preprint to Muhan Gao and coauthors on arXiv cs.AI. Findings and the design principle are consistently framed as the authors' reports and arguments, not established facts.","excerpt":"Authors:Muhan Gao, Hangxiao Zhu, Xinqi Su, Yu Zhang, Daniel Khashabi, Kuan-Hao Huang"},{"dimension":"summary","status":"pass","explanation":"The reading faithfully covers the abstract's setup, scale, early-progress plateau, generation-versus-judgment bottleneck, the two failure modes and the proposed design principle. The yes-or-no game is explained in plain language. Terms such as binary oracle feedback and autoresearch are kept exactly, consistent with term-library tooltips. The question follows from the evidence. A minor omission is that the first 30 questions 'largely decide how the game ends', but this does not distort the meaning.","excerpt":"nearly all progress happens within the first 30 of 100 questions, which largely decide how the game ends"},{"dimension":"date","status":"pass","explanation":"The value of originalPublished, 2026-11-10, matches the arXiv v1 submission date. The feed date of 2026-11-11 is correctly not used.","excerpt":"[Submitted on 10 Nov 2026]"},{"dimension":"support","status":"pass","explanation":"Each claim is directly supported by the abstract. The design principle is correctly labeled as a proposal rather than a demonstrated improvement, preserving the source's hedged 'may be more effective'. The limitations accurately note that only the abstract is available. No unsupported benefits, costs or forecasts are added.","excerpt":"enriching the search with externally sourced ideas and questions may be more effective than improving any single model's generation ability"},{"dimension":"media","status":"pass","explanation":"The selected media is the verified direct link to the arXiv abstract page. It is the original source and the most direct access point.","excerpt":null}],"createdAt":"2026-11-12T02:17:11.268Z","model":"anthropic/claude-opus-5.5"}],"correctionUrl":"https://escapemeridian.com/corrections/"}