{"id":"1aa06cf2522d2fcdd9da74494d12f5926f38986bbcc609f159ab7bf53994b1c5","title":"Jusheng Zhang, Keze Wang and Liang Lin: Can LLMs Debug Robots by Editing Their Weights?","argument":"Jusheng Zhang, Keze Wang and Liang Lin report that a language model can repair a robot policy’s lexical instruction grounding by editing 65 token-embedding rows. In this approach, GPT-6 Astra reads the policy’s weights, works in a CPU workspace and uses feedback from a simulator, without gradient updates or new demonstrations. Across three independent sessions on π0.5, all edits improved success on 765 held-out paraphrases; the best raised success from 86.3% to 92.0%, with 48 repairs and four regressions. The gains were concentrated in instructions containing edited tokens: success rose from 74.3% to 87.5%, while the other episodes had no outcome changes. Matched random-direction edits did not help. In 13 further robot sessions, the editor declined 12 candidates, and the one retained edit showed no significant held-out gain. The authors also report that benchmark metadata reduced FastWAM success, while the evaluated VLA’s aggregate rate was unchanged. They argue that the results support precise lexical repair, not general acquisition of missing robot capabilities.","topicIds":["intelligence-and-discovery"],"domain":"science","claimType":"attributed reading","limits":"The abstract’s main improvement results concern one robot policy, π0.5, and held-out paraphrases; it does not establish general acquisition of missing robot capabilities. In 13 further robot sessions, only one candidate was retained, and that edit did not show a significant held-out gain.","speaker":{"name":"Zhang, Jusheng, Wang, Keze, Lin, Liang","roleAtSource":"Article author"},"source":{"title":"Can LLMs Debug Robots by Editing Their Weights?","publisher":"arXiv cs.AI","format":"Article","url":"https://arxiv.org/abs/2611.09376","publishedAt":"2026-11-06","dateBasis":"Date basis not provided."},"evidence":[{"text":"The authors report that a language model repaired a robot policy’s lexical instruction grounding by editing 65 token-embedding rows, without gradient updates or new demonstrations.","excerpt":"A language model can repair a robot policy's lexical instruction grounding by editing just 65 token-embedding rows.","url":"https://arxiv.org/abs/2611.09376","relationship":"Source-stated"},{"text":"Across three independent sessions on π0.5, all edits improved success on 765 held-out paraphrases; the best result increased success from 86.3% to 92.0%, with 48 repairs and four regressions.","excerpt":"Three independent sessions on $\\pi_{0.5}$ all improve success on 765 held-out paraphrases; the best raises success from 86.3\\% to 92.0\\% (48 repairs, four regressions;","url":"https://arxiv.org/abs/2611.09376","relationship":"Source-stated"},{"text":"The reported gains were limited to instructions containing edited tokens; the other 430 episodes had no outcome changes.","excerpt":"Among the 335 instructions containing edited tokens, success rises from 74.3\\% to 87.5\\%; the other 430 episodes have zero outcome changes.","url":"https://arxiv.org/abs/2611.09376","relationship":"Source-stated"},{"text":"Matched random-direction edits did not help, and in 13 further robot sessions the editor retained only one candidate, which showed no significant held-out gain.","excerpt":"Matched random-direction edits do not help. Full replacement using the editor's own lexicon explains much of the gain but produces more regressions; its direct difference from the editor is not significant at 0.05. Across 13 further robot sessions, the editor declines 12 candidates; the sole","url":"https://arxiv.org/abs/2611.09376","relationship":"Source-stated"},{"text":"The authors characterize the evidence as supporting precise lexical repair, rather than general acquisition of missing robot capabilities.","excerpt":"The evidence supports precise lexical repair, not general acquisition of missing robot capabilities.","url":"https://arxiv.org/abs/2611.09376","relationship":"Source-stated"}],"sourceMedia":{"status":"not_found","sourceUrl":"https://arxiv.org/abs/2611.09376","checkedAt":"2026-11-13T01:02:09.054Z","items":[],"note":"No suitable media was found by the supported page extraction."},"assumptions":[],"openQuestion":{"attribution":"Open question","text":"How would the reported lexical repair perform on robot policies beyond π0.5?"},"whatWouldChangeView":{"attribution":"Change condition not recorded","text":""},"editorialComparisons":[],"live":{"title":"Jusheng Zhang, Keze Wang and Liang Lin: Can LLMs Debug Robots by Editing Their Weights?","summary":"Jusheng Zhang, Keze Wang and Liang Lin report that a language model can repair a robot policy’s lexical instruction grounding by editing 65 token-embedding rows. In this approach, GPT-6 Astra reads the policy’s weights, works in a CPU workspace and uses feedback from a simulator, without gradient updates or new demonstrations. Across three independent sessions on π0.5, all edits improved success on 765 held-out paraphrases; the best raised success from 86.3% to 92.0%, with 48 repairs and four regressions. The gains were concentrated in instructions containing edited tokens: success rose from 74.3% to 87.5%, while the other episodes had no outcome changes. Matched random-direction edits did not help. In 13 further robot sessions, the editor declined 12 candidates, and the one retained edit showed no significant held-out gain. The authors also report that benchmark metadata reduced FastWAM success, while the evaluated VLA’s aggregate rate was unchanged. They argue that the results support precise lexical repair, not general acquisition of missing robot capabilities.","question":"How would the reported lexical repair perform on robot policies beyond π0.5?","claims":[{"text":"The authors report that a language model repaired a robot policy’s lexical instruction grounding by editing 65 token-embedding rows, without gradient updates or new demonstrations.","excerpt":"A language model can repair a robot policy's lexical instruction grounding by editing just 65 token-embedding rows.","support":"source-stated"},{"text":"Across three independent sessions on π0.5, all edits improved success on 765 held-out paraphrases; the best result increased success from 86.3% to 92.0%, with 48 repairs and four regressions.","excerpt":"Three independent sessions on $\\pi_{0.5}$ all improve success on 765 held-out paraphrases; the best raises success from 86.3\\% to 92.0\\% (48 repairs, four regressions;","support":"source-stated"},{"text":"The reported gains were limited to instructions containing edited tokens; the other 430 episodes had no outcome changes.","excerpt":"Among the 335 instructions containing edited tokens, success rises from 74.3\\% to 87.5\\%; the other 430 episodes have zero outcome changes.","support":"source-stated"},{"text":"Matched random-direction edits did not help, and in 13 further robot sessions the editor retained only one candidate, which showed no significant held-out gain.","excerpt":"Matched random-direction edits do not help. Full replacement using the editor's own lexicon explains much of the gain but produces more regressions; its direct difference from the editor is not significant at 0.05. Across 13 further robot sessions, the editor declines 12 candidates; the sole","support":"source-stated"},{"text":"The authors characterize the evidence as supporting precise lexical repair, rather than general acquisition of missing robot capabilities.","excerpt":"The evidence supports precise lexical repair, not general acquisition of missing robot capabilities.","support":"source-stated"}],"limitations":["The abstract’s main improvement results concern one robot policy, π0.5, and held-out paraphrases; it does not establish general acquisition of missing robot capabilities.","In 13 further robot sessions, only one candidate was retained, and that edit did not show a significant held-out gain."],"mediaId":"9ab1d833873294bc639e9d805e6ce9f98a62266660cfc5be68a86c28be37fde8","correctionReason":null,"id":"1aa06cf2522d2fcdd9da74494d12f5926f38986bbcc609f159ab7bf53994b1c5","publicationId":"fcce4db2-cfe4-4d48-8377-82c41276b584","version":1,"sourceUrl":"https://arxiv.org/abs/2611.09376","sourceTitle":"Can LLMs Debug Robots by Editing Their Weights?","publisher":"arXiv cs.AI","publishedAt":null,"originalPublished":{"value":"2026-11-06","precision":"day","basis":"publisher_citation_date"},"firstObservedAt":"2026-11-09T05:00:27.699Z","createdAt":"2026-11-13T00:31:40.790Z","historical":true,"entityIds":["arxiv"],"topicIds":["intelligence-and-discovery"],"media":[{"id":"9ab1d833873294bc639e9d805e6ce9f98a62266660cfc5be68a86c28be37fde8","kind":"article","url":"https://arxiv.org/abs/2611.09376","sourceUrl":"https://arxiv.org/abs/2611.09376","label":"Read original","provenance":"source_link","verification":{"status":"verified","checkedAt":"2026-11-13T00:30:48.826Z","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/2611.09376","checkedAt":"2026-11-13T01:02:09.054Z","items":[],"note":"No suitable media was found by the supported page extraction."},"editorialDecision":{"outcome":"accept","reasonCodes":["material_development"],"rubricVersion":"stream-admission-v6","mediaId":"9ab1d833873294bc639e9d805e6ce9f98a62266660cfc5be68a86c28be37fde8","confidence":0.68,"provider":"typesafe","model":"jev-1.13.0","inputHash":"c38ce483a18d1830641037849b0923b6069cd356fcfa75d6ad4374cf2c46f87c","decidedAt":"2026-11-13T00:30:50.056Z"},"provenance":{"textHash":"d853023055e30b3b4817b8da4b551aa12edc99a37bdb13149877a426ca5c4008","observationId":"06dfb824a0222f5fae365ab14fcb48f4f0a6f7b8d7b85b53e5ac1aa75cd24222","retrievedAt":"2026-11-13T00:30:48.697Z","kind":"publisher-page","preparer":{"model":"gpt-6-luna","runId":"b5034653-9151-46af-8ae1-0c3fcae219c6"},"reviewer":{"model":"gpt-6-luna","runId":"b5a1ff04-9cbf-4a5b-8bde-c0991d51ba3c"}},"audit":{"status":"uncertain","reviewedAt":"2026-11-13T00:31:40.790Z","model":"anthropic/claude-opus-5.5"},"sourceAttribution":{"authors":[{"name":"Zhang, Jusheng","url":null,"kind":"unknown"},{"name":"Wang, Keze","url":null,"kind":"unknown"},{"name":"Lin, Liang","url":null,"kind":"unknown"}],"authorRole":"article-author","hosts":[],"status":"stated","authorRoster":{"truncated":false,"omitted":0,"assertionGroupsOmitted":0},"participation":[{"actor":{"name":"Zhang, Jusheng","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Zhang, Jusheng","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2611.09376","locator":"head/meta[name=author|citation_author]"},"assertionId":"c1ce5f3f0cda324ba13cd27b6410b9f59ad10c54e7c994af3181e2e06b874fa2:participation:0","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-13T01:02:09.054Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"c1ce5f3f0cda324ba13cd27b6410b9f59ad10c54e7c994af3181e2e06b874fa2","sourceUrl":"https://arxiv.org/abs/2611.09376","contentSha256":"35d08c68dbfb94986a74573d3b7c9e247984bf3f2a0fb4210c832b9a2e0941c5","textHash":"d853023055e30b3b4817b8da4b551aa12edc99a37bdb13149877a426ca5c4008","observedAt":"2026-11-13T01:02:09.054Z"}},{"actor":{"name":"Wang, Keze","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Wang, Keze","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2611.09376","locator":"head/meta[name=author|citation_author]"},"assertionId":"c1ce5f3f0cda324ba13cd27b6410b9f59ad10c54e7c994af3181e2e06b874fa2:participation:1","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-13T01:02:09.054Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"c1ce5f3f0cda324ba13cd27b6410b9f59ad10c54e7c994af3181e2e06b874fa2","sourceUrl":"https://arxiv.org/abs/2611.09376","contentSha256":"35d08c68dbfb94986a74573d3b7c9e247984bf3f2a0fb4210c832b9a2e0941c5","textHash":"d853023055e30b3b4817b8da4b551aa12edc99a37bdb13149877a426ca5c4008","observedAt":"2026-11-13T01:02:09.054Z"}},{"actor":{"name":"Lin, Liang","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Lin, Liang","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2611.09376","locator":"head/meta[name=author|citation_author]"},"assertionId":"c1ce5f3f0cda324ba13cd27b6410b9f59ad10c54e7c994af3181e2e06b874fa2:participation:2","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-13T01:02:09.054Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"c1ce5f3f0cda324ba13cd27b6410b9f59ad10c54e7c994af3181e2e06b874fa2","sourceUrl":"https://arxiv.org/abs/2611.09376","contentSha256":"35d08c68dbfb94986a74573d3b7c9e247984bf3f2a0fb4210c832b9a2e0941c5","textHash":"d853023055e30b3b4817b8da4b551aa12edc99a37bdb13149877a426ca5c4008","observedAt":"2026-11-13T01:02:09.054Z"}}],"participationCoverage":{"total":3,"returned":3,"complete":true,"fullAssertions":"source-metadata-endpoint"},"distribution":null,"publisher":null,"sourceUrl":"https://arxiv.org/abs/2611.09376","snapshotId":"c1ce5f3f0cda324ba13cd27b6410b9f59ad10c54e7c994af3181e2e06b874fa2","contentSha256":"35d08c68dbfb94986a74573d3b7c9e247984bf3f2a0fb4210c832b9a2e0941c5","textHash":"d853023055e30b3b4817b8da4b551aa12edc99a37bdb13149877a426ca5c4008","observedAt":"2026-11-13T01:02:09.054Z","basis":"publisher-page-metadata","sourceLinks":42,"linksTruncated":false,"metadataUrl":"/api/pipeline/stream/1aa06cf2522d2fcdd9da74494d12f5926f38986bbcc609f159ab7bf53994b1c5/source"}},"canonicalUrl":"https://escapemeridian.com/readings/jusheng-zhang-keze-wang-and-liang-lin-can-llms-debug-robots-by-1aa06cf2","formats":{"html":"/readings/jusheng-zhang-keze-wang-and-liang-lin-can-llms-debug-robots-by-1aa06cf2","markdown":"/readings/jusheng-zhang-keze-wang-and-liang-lin-can-llms-debug-robots-by-1aa06cf2.md","json":"/readings/jusheng-zhang-keze-wang-and-liang-lin-can-llms-debug-robots-by-1aa06cf2.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"},{"id":"term-cpu","name":"CPU","explanation":"The central processing unit: a general-purpose processor that carries out a computer's instructions.","referenceUrl":"https://aws.amazon.com/what-is/cpu/"},{"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"}],"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":"1aa06cf2522d2fcdd9da74494d12f5926f38986bbcc609f159ab7bf53994b1c5","revision":"fcce4db2-cfe4-4d48-8377-82c41276b584","revisionKind":"publication-id","title":"Jusheng Zhang, Keze Wang and Liang Lin: Can LLMs Debug Robots by Editing Their Weights?","url":"https://escapemeridian.com/readings/jusheng-zhang-keze-wang-and-liang-lin-can-llms-debug-robots-by-1aa06cf2"},"source":{"type":"source","id":"https://arxiv.org/abs/2611.09376","revision":"d853023055e30b3b4817b8da4b551aa12edc99a37bdb13149877a426ca5c4008","revisionKind":"sha256-source-text","title":"Can LLMs Debug Robots by Editing Their Weights?","url":"https://arxiv.org/abs/2611.09376"},"status":"partial","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":"1aa06cf2522d2fcdd9da74494d12f5926f38986bbcc609f159ab7bf53994b1c5","title":"Jusheng Zhang, Keze Wang and Liang Lin: Can LLMs Debug Robots by Editing Their Weights?","argument":"Jusheng Zhang, Keze Wang and Liang Lin report that a language model can repair a robot policy’s lexical instruction grounding by editing 65 token-embedding rows. In this approach, GPT-6 Astra reads the policy’s weights, works in a CPU workspace and uses feedback from a simulator, without gradient updates or new demonstrations. Across three independent sessions on π0.5, all edits improved success on 765 held-out paraphrases; the best raised success from 86.3% to 92.0%, with 48 repairs and four regressions. The gains were concentrated in instructions containing edited tokens: success rose from 74.3% to 87.5%, while the other episodes had no outcome changes. Matched random-direction edits did not help. In 13 further robot sessions, the editor declined 12 candidates, and the one retained edit showed no significant held-out gain. The authors also report that benchmark metadata reduced FastWAM success, while the evaluated VLA’s aggregate rate was unchanged. They argue that the results support precise lexical repair, not general acquisition of missing robot capabilities.","topicIds":["intelligence-and-discovery"],"domain":"science","claimType":"attributed reading","limits":"The abstract’s main improvement results concern one robot policy, π0.5, and held-out paraphrases; it does not establish general acquisition of missing robot capabilities. In 13 further robot sessions, only one candidate was retained, and that edit did not show a significant held-out gain.","speaker":{"name":"Zhang, Jusheng, Wang, Keze, Lin, Liang","roleAtSource":"Article author"},"source":{"title":"Can LLMs Debug Robots by Editing Their Weights?","publisher":"arXiv cs.AI","format":"Article","url":"https://arxiv.org/abs/2611.09376","publishedAt":"2026-11-06","dateBasis":"Date basis not provided."},"evidence":[{"text":"The authors report that a language model repaired a robot policy’s lexical instruction grounding by editing 65 token-embedding rows, without gradient updates or new demonstrations.","excerpt":"A language model can repair a robot policy's lexical instruction grounding by editing just 65 token-embedding rows.","url":"https://arxiv.org/abs/2611.09376","relationship":"Source-stated"},{"text":"Across three independent sessions on π0.5, all edits improved success on 765 held-out paraphrases; the best result increased success from 86.3% to 92.0%, with 48 repairs and four regressions.","excerpt":"Three independent sessions on $\\pi_{0.5}$ all improve success on 765 held-out paraphrases; the best raises success from 86.3\\% to 92.0\\% (48 repairs, four regressions;","url":"https://arxiv.org/abs/2611.09376","relationship":"Source-stated"},{"text":"The reported gains were limited to instructions containing edited tokens; the other 430 episodes had no outcome changes.","excerpt":"Among the 335 instructions containing edited tokens, success rises from 74.3\\% to 87.5\\%; the other 430 episodes have zero outcome changes.","url":"https://arxiv.org/abs/2611.09376","relationship":"Source-stated"},{"text":"Matched random-direction edits did not help, and in 13 further robot sessions the editor retained only one candidate, which showed no significant held-out gain.","excerpt":"Matched random-direction edits do not help. Full replacement using the editor's own lexicon explains much of the gain but produces more regressions; its direct difference from the editor is not significant at 0.05. Across 13 further robot sessions, the editor declines 12 candidates; the sole","url":"https://arxiv.org/abs/2611.09376","relationship":"Source-stated"},{"text":"The authors characterize the evidence as supporting precise lexical repair, rather than general acquisition of missing robot capabilities.","excerpt":"The evidence supports precise lexical repair, not general acquisition of missing robot capabilities.","url":"https://arxiv.org/abs/2611.09376","relationship":"Source-stated"}],"sourceMedia":{"status":"not_found","sourceUrl":"https://arxiv.org/abs/2611.09376","checkedAt":"2026-11-13T01:02:09.054Z","items":[],"note":"No suitable media was found by the supported page extraction."},"assumptions":[],"openQuestion":{"attribution":"Open question","text":"How would the reported lexical repair perform on robot policies beyond π0.5?"},"whatWouldChangeView":{"attribution":"Change condition not recorded","text":""},"editorialComparisons":[],"live":{"title":"Jusheng Zhang, Keze Wang and Liang Lin: Can LLMs Debug Robots by Editing Their Weights?","summary":"Jusheng Zhang, Keze Wang and Liang Lin report that a language model can repair a robot policy’s lexical instruction grounding by editing 65 token-embedding rows. In this approach, GPT-6 Astra reads the policy’s weights, works in a CPU workspace and uses feedback from a simulator, without gradient updates or new demonstrations. Across three independent sessions on π0.5, all edits improved success on 765 held-out paraphrases; the best raised success from 86.3% to 92.0%, with 48 repairs and four regressions. The gains were concentrated in instructions containing edited tokens: success rose from 74.3% to 87.5%, while the other episodes had no outcome changes. Matched random-direction edits did not help. In 13 further robot sessions, the editor declined 12 candidates, and the one retained edit showed no significant held-out gain. The authors also report that benchmark metadata reduced FastWAM success, while the evaluated VLA’s aggregate rate was unchanged. They argue that the results support precise lexical repair, not general acquisition of missing robot capabilities.","question":"How would the reported lexical repair perform on robot policies beyond π0.5?","claims":[{"text":"The authors report that a language model repaired a robot policy’s lexical instruction grounding by editing 65 token-embedding rows, without gradient updates or new demonstrations.","excerpt":"A language model can repair a robot policy's lexical instruction grounding by editing just 65 token-embedding rows.","support":"source-stated"},{"text":"Across three independent sessions on π0.5, all edits improved success on 765 held-out paraphrases; the best result increased success from 86.3% to 92.0%, with 48 repairs and four regressions.","excerpt":"Three independent sessions on $\\pi_{0.5}$ all improve success on 765 held-out paraphrases; the best raises success from 86.3\\% to 92.0\\% (48 repairs, four regressions;","support":"source-stated"},{"text":"The reported gains were limited to instructions containing edited tokens; the other 430 episodes had no outcome changes.","excerpt":"Among the 335 instructions containing edited tokens, success rises from 74.3\\% to 87.5\\%; the other 430 episodes have zero outcome changes.","support":"source-stated"},{"text":"Matched random-direction edits did not help, and in 13 further robot sessions the editor retained only one candidate, which showed no significant held-out gain.","excerpt":"Matched random-direction edits do not help. Full replacement using the editor's own lexicon explains much of the gain but produces more regressions; its direct difference from the editor is not significant at 0.05. Across 13 further robot sessions, the editor declines 12 candidates; the sole","support":"source-stated"},{"text":"The authors characterize the evidence as supporting precise lexical repair, rather than general acquisition of missing robot capabilities.","excerpt":"The evidence supports precise lexical repair, not general acquisition of missing robot capabilities.","support":"source-stated"}],"limitations":["The abstract’s main improvement results concern one robot policy, π0.5, and held-out paraphrases; it does not establish general acquisition of missing robot capabilities.","In 13 further robot sessions, only one candidate was retained, and that edit did not show a significant held-out gain."],"mediaId":"9ab1d833873294bc639e9d805e6ce9f98a62266660cfc5be68a86c28be37fde8","correctionReason":null,"id":"1aa06cf2522d2fcdd9da74494d12f5926f38986bbcc609f159ab7bf53994b1c5","publicationId":"fcce4db2-cfe4-4d48-8377-82c41276b584","version":1,"sourceUrl":"https://arxiv.org/abs/2611.09376","sourceTitle":"Can LLMs Debug Robots by Editing Their Weights?","publisher":"arXiv cs.AI","publishedAt":null,"originalPublished":{"value":"2026-11-06","precision":"day","basis":"publisher_citation_date"},"firstObservedAt":"2026-11-09T05:00:27.699Z","createdAt":"2026-11-13T00:31:40.790Z","historical":true,"entityIds":["arxiv"],"topicIds":["intelligence-and-discovery"],"media":[{"id":"9ab1d833873294bc639e9d805e6ce9f98a62266660cfc5be68a86c28be37fde8","kind":"article","url":"https://arxiv.org/abs/2611.09376","sourceUrl":"https://arxiv.org/abs/2611.09376","label":"Read original","provenance":"source_link","verification":{"status":"verified","checkedAt":"2026-11-13T00:30:48.826Z","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/2611.09376","checkedAt":"2026-11-13T01:02:09.054Z","items":[],"note":"No suitable media was found by the supported page extraction."},"editorialDecision":{"outcome":"accept","reasonCodes":["material_development"],"rubricVersion":"stream-admission-v6","mediaId":"9ab1d833873294bc639e9d805e6ce9f98a62266660cfc5be68a86c28be37fde8","confidence":0.68,"provider":"typesafe","model":"jev-1.13.0","inputHash":"c38ce483a18d1830641037849b0923b6069cd356fcfa75d6ad4374cf2c46f87c","decidedAt":"2026-11-13T00:30:50.056Z"},"provenance":{"textHash":"d853023055e30b3b4817b8da4b551aa12edc99a37bdb13149877a426ca5c4008","observationId":"06dfb824a0222f5fae365ab14fcb48f4f0a6f7b8d7b85b53e5ac1aa75cd24222","retrievedAt":"2026-11-13T00:30:48.697Z","kind":"publisher-page","preparer":{"model":"gpt-6-luna","runId":"b5034653-9151-46af-8ae1-0c3fcae219c6"},"reviewer":{"model":"gpt-6-luna","runId":"b5a1ff04-9cbf-4a5b-8bde-c0991d51ba3c"}},"audit":{"status":"uncertain","reviewedAt":"2026-11-13T00:31:40.790Z","model":"anthropic/claude-opus-5.5"},"sourceAttribution":{"authors":[{"name":"Zhang, Jusheng","url":null,"kind":"unknown"},{"name":"Wang, Keze","url":null,"kind":"unknown"},{"name":"Lin, Liang","url":null,"kind":"unknown"}],"authorRole":"article-author","hosts":[],"status":"stated","authorRoster":{"truncated":false,"omitted":0,"assertionGroupsOmitted":0},"participation":[{"actor":{"name":"Zhang, Jusheng","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Zhang, Jusheng","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2611.09376","locator":"head/meta[name=author|citation_author]"},"assertionId":"c1ce5f3f0cda324ba13cd27b6410b9f59ad10c54e7c994af3181e2e06b874fa2:participation:0","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-13T01:02:09.054Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"c1ce5f3f0cda324ba13cd27b6410b9f59ad10c54e7c994af3181e2e06b874fa2","sourceUrl":"https://arxiv.org/abs/2611.09376","contentSha256":"35d08c68dbfb94986a74573d3b7c9e247984bf3f2a0fb4210c832b9a2e0941c5","textHash":"d853023055e30b3b4817b8da4b551aa12edc99a37bdb13149877a426ca5c4008","observedAt":"2026-11-13T01:02:09.054Z"}},{"actor":{"name":"Wang, Keze","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Wang, Keze","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2611.09376","locator":"head/meta[name=author|citation_author]"},"assertionId":"c1ce5f3f0cda324ba13cd27b6410b9f59ad10c54e7c994af3181e2e06b874fa2:participation:1","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-13T01:02:09.054Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"c1ce5f3f0cda324ba13cd27b6410b9f59ad10c54e7c994af3181e2e06b874fa2","sourceUrl":"https://arxiv.org/abs/2611.09376","contentSha256":"35d08c68dbfb94986a74573d3b7c9e247984bf3f2a0fb4210c832b9a2e0941c5","textHash":"d853023055e30b3b4817b8da4b551aa12edc99a37bdb13149877a426ca5c4008","observedAt":"2026-11-13T01:02:09.054Z"}},{"actor":{"name":"Lin, Liang","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Lin, Liang","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2611.09376","locator":"head/meta[name=author|citation_author]"},"assertionId":"c1ce5f3f0cda324ba13cd27b6410b9f59ad10c54e7c994af3181e2e06b874fa2:participation:2","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-13T01:02:09.054Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"c1ce5f3f0cda324ba13cd27b6410b9f59ad10c54e7c994af3181e2e06b874fa2","sourceUrl":"https://arxiv.org/abs/2611.09376","contentSha256":"35d08c68dbfb94986a74573d3b7c9e247984bf3f2a0fb4210c832b9a2e0941c5","textHash":"d853023055e30b3b4817b8da4b551aa12edc99a37bdb13149877a426ca5c4008","observedAt":"2026-11-13T01:02:09.054Z"}}],"participationCoverage":{"total":3,"returned":3,"complete":true,"fullAssertions":"source-metadata-endpoint"},"distribution":null,"publisher":null,"sourceUrl":"https://arxiv.org/abs/2611.09376","snapshotId":"c1ce5f3f0cda324ba13cd27b6410b9f59ad10c54e7c994af3181e2e06b874fa2","contentSha256":"35d08c68dbfb94986a74573d3b7c9e247984bf3f2a0fb4210c832b9a2e0941c5","textHash":"d853023055e30b3b4817b8da4b551aa12edc99a37bdb13149877a426ca5c4008","observedAt":"2026-11-13T01:02:09.054Z","basis":"publisher-page-metadata","sourceLinks":42,"linksTruncated":false,"metadataUrl":"/api/pipeline/stream/1aa06cf2522d2fcdd9da74494d12f5926f38986bbcc609f159ab7bf53994b1c5/source"}}}],"audits":[{"publicationId":"fcce4db2-cfe4-4d48-8377-82c41276b584","verdict":"uncertain","findings":[{"dimension":"attribution","status":"pass","explanation":"The reading correctly attributes the paper to Jusheng Zhang, Keze Wang and Liang Lin, and presents the findings as the authors' reports rather than as established fact.","excerpt":"Authors:Jusheng Zhang, Keze Wang, Liang Lin"},{"dimension":"summary","status":"uncertain","explanation":"Two problems. First, the summary omits the abstract's key baseline: full replacement using the editor's own lexicon explains much of the gain, and its difference from the editor is not significant. Without this, the reader may credit the editor more than the source does. Repair: add one sentence on that baseline. Second, central terms are unexplained for a general reader: lexical instruction grounding, token-embedding rows, gradient updates, π0.5, VLA, FastWAM and benchmark metadata. The FastWAM sentence is especially opaque. Repair: use term-library tooltips and add plain-language framing.","excerpt":"Full replacement using the editor's own lexicon explains much of the gain but produces more regressions; its direct difference from the editor is not significant at 0.05."},{"dimension":"date","status":"pass","explanation":"The displayed date of 2026-11-06 matches the v1 submission date. The later feed date is correctly not used.","excerpt":"[Submitted on 6 Nov 2026]"},{"dimension":"support","status":"pass","explanation":"The claims, the limitations and the question about other policies all follow from the abstract. One minor wording issue: 'all edits improved' should read 'all three sessions improved', which is the source's framing. The fourth claim's excerpt includes the full-replacement caveat, but the claim text drops it. That omission is addressed under summary.","excerpt":"Three independent sessions on $\\pi_{0.5}$ all improve success on 765 held-out paraphrases"},{"dimension":"media","status":"pass","explanation":"The selected media is the verified direct arXiv abstract page for this paper, which is the most direct way to access the source.","excerpt":null}],"createdAt":"2026-11-13T00:31:40.790Z","model":"anthropic/claude-opus-5.5"}],"correctionUrl":"https://escapemeridian.com/corrections/"}