{"id":"b9a32c33bd8c9037a013d311d5b9f3285af49b9fa6600368ed43b33d4b81ea44","title":"Gated DeltaNet-2: Decoupling Erase and Write in Linear Attention","argument":"Ali Hatamizadeh, Yejin Choi and Jan Kautz introduce Gated DeltaNet-2, a linear attention model that separates two operations in its compressed recurrent memory: erasing old content and writing new content. They argue that earlier delta-rule models tied these operations to one scalar gate, limiting how memory could be edited. Their method uses separate channel-wise erase and write gates, while retaining adaptive forgetting and channel-wise decay. The paper also describes a chunkwise WY algorithm and a gate-aware backward pass for parallel training. In experiments with a 1.3B-parameter model trained on 100B FineWeb-Edu tokens, the authors report the strongest overall results among the compared Mamba-2, Gated DeltaNet, KDA and Mamba-3 variants across language modeling, commonsense reasoning and retrieval. They say the advantage is most pronounced on long-context RULER needle-in-a-haystack benchmarks, particularly multi-key retrieval, and report strong results in recurrent and hybrid settings. These findings concern the reported model size, training data and evaluations; they do not establish performance in other settings.","topicIds":["intelligence-and-discovery"],"domain":"science","claimType":"attributed reading","limits":"The reported results are tied to the stated model size, training data, compared variants and evaluations; the evidence does not establish performance beyond those settings. The source gives aggregate claims about benchmark results but does not provide numerical scores in the supplied abstract.","speaker":{"name":"Hatamizadeh, Ali, Choi, Yejin, Kautz, Jan","roleAtSource":"Article author"},"source":{"title":"Gated DeltaNet-2: Decoupling Erase and Write in Linear Attention","publisher":"arXiv cs.AI","format":"Article","url":"https://arxiv.org/abs/2605.22791","publishedAt":"2026-05-21","dateBasis":"Date basis not provided."},"evidence":[{"text":"The paper introduces separate channel-wise erase and write gates to address the scalar tie between erasing old content and committing new content.","excerpt":"the active edit still uses a single scalar gate to control two different things: how much old content to erase on the key side and how much new content to commit on the value side.","url":"https://arxiv.org/abs/2605.22791","relationship":"Source-stated"},{"text":"The authors report the strongest overall results among the compared model variants across three task areas in their stated training setup.","excerpt":"At 1.3B parameters trained on 100B FineWeb-Edu tokens, Gated DeltaNet-2 achieves the strongest overall results among Mamba-2, Gated DeltaNet, KDA, and Mamba-3 variants across language modeling, commonsense reasoning, and retrieval.","url":"https://arxiv.org/abs/2605.22791","relationship":"Source-stated"},{"text":"The authors report the clearest advantage on long-context RULER needle-in-a-haystack benchmarks, including multi-key retrieval.","excerpt":"Its advantage is most pronounced on long-context RULER needle-in-a-haystack benchmarks, where it improves the evaluated multi-key retrieval setting and remains strong in both recurrent and hybrid settings.","url":"https://arxiv.org/abs/2605.22791","relationship":"Source-stated"}],"sourceMedia":{"status":"not_found","sourceUrl":"https://arxiv.org/abs/2605.22791","checkedAt":"2026-11-11T11:54:06.837Z","items":[],"note":"No suitable media was found by the supported page extraction."},"assumptions":[],"openQuestion":{"attribution":"Open question","text":"How would Gated DeltaNet-2 perform on long-context retrieval beyond the evaluated RULER settings?"},"whatWouldChangeView":{"attribution":"Change condition not recorded","text":""},"editorialComparisons":[],"live":{"title":"Gated DeltaNet-2: Decoupling Erase and Write in Linear Attention","summary":"Ali Hatamizadeh, Yejin Choi and Jan Kautz introduce Gated DeltaNet-2, a linear attention model that separates two operations in its compressed recurrent memory: erasing old content and writing new content. They argue that earlier delta-rule models tied these operations to one scalar gate, limiting how memory could be edited. Their method uses separate channel-wise erase and write gates, while retaining adaptive forgetting and channel-wise decay. The paper also describes a chunkwise WY algorithm and a gate-aware backward pass for parallel training. In experiments with a 1.3B-parameter model trained on 100B FineWeb-Edu tokens, the authors report the strongest overall results among the compared Mamba-2, Gated DeltaNet, KDA and Mamba-3 variants across language modeling, commonsense reasoning and retrieval. They say the advantage is most pronounced on long-context RULER needle-in-a-haystack benchmarks, particularly multi-key retrieval, and report strong results in recurrent and hybrid settings. These findings concern the reported model size, training data and evaluations; they do not establish performance in other settings.","question":"How would Gated DeltaNet-2 perform on long-context retrieval beyond the evaluated RULER settings?","claims":[{"text":"The paper introduces separate channel-wise erase and write gates to address the scalar tie between erasing old content and committing new content.","excerpt":"the active edit still uses a single scalar gate to control two different things: how much old content to erase on the key side and how much new content to commit on the value side.","support":"source-stated"},{"text":"The authors report the strongest overall results among the compared model variants across three task areas in their stated training setup.","excerpt":"At 1.3B parameters trained on 100B FineWeb-Edu tokens, Gated DeltaNet-2 achieves the strongest overall results among Mamba-2, Gated DeltaNet, KDA, and Mamba-3 variants across language modeling, commonsense reasoning, and retrieval.","support":"source-stated"},{"text":"The authors report the clearest advantage on long-context RULER needle-in-a-haystack benchmarks, including multi-key retrieval.","excerpt":"Its advantage is most pronounced on long-context RULER needle-in-a-haystack benchmarks, where it improves the evaluated multi-key retrieval setting and remains strong in both recurrent and hybrid settings.","support":"source-stated"}],"limitations":["The reported results are tied to the stated model size, training data, compared variants and evaluations; the evidence does not establish performance beyond those settings.","The source gives aggregate claims about benchmark results but does not provide numerical scores in the supplied abstract."],"mediaId":"93b69d67e028310f9beb9feca15977b198c6f8607f601e6fb23438e6ada8baca","correctionReason":null,"id":"b9a32c33bd8c9037a013d311d5b9f3285af49b9fa6600368ed43b33d4b81ea44","publicationId":"51a904f3-8e05-40e5-82d3-fd9d2e47be54","version":1,"sourceUrl":"https://arxiv.org/abs/2605.22791","sourceTitle":"Gated DeltaNet-2: Decoupling Erase and Write in Linear Attention","publisher":"arXiv cs.AI","publishedAt":null,"originalPublished":{"value":"2026-05-21","precision":"day","basis":"publisher_citation_date"},"firstObservedAt":"2026-11-09T05:04:08.699Z","createdAt":"2026-11-10T18:50:12.923Z","historical":true,"entityIds":["arxiv"],"topicIds":["intelligence-and-discovery"],"media":[{"id":"93b69d67e028310f9beb9feca15977b198c6f8607f601e6fb23438e6ada8baca","kind":"article","url":"https://arxiv.org/abs/2605.22791","sourceUrl":"https://arxiv.org/abs/2605.22791","label":"Read original","provenance":"source_link","verification":{"status":"verified","checkedAt":"2026-11-09T05:32:30.649Z","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/2605.22791","checkedAt":"2026-11-11T11:54:06.837Z","items":[],"note":"No suitable media was found by the supported page extraction."},"editorialDecision":{"outcome":"accept","reasonCodes":["material_development"],"rubricVersion":"stream-admission-v5","mediaId":"93b69d67e028310f9beb9feca15977b198c6f8607f601e6fb23438e6ada8baca","confidence":0.93,"provider":"typesafe","model":"jev-1.13.0","inputHash":"4eedfd150e1d66243dc6f4aaf20f6f097fc64de1ada5ec899664d1211d4ccde3","decidedAt":"2026-11-09T05:32:32.553Z"},"provenance":{"textHash":"f3f6dd0e01bc93c2b075466ec451e7ca79310c41dda70b6388f37c48777fca9b","observationId":"7b1143e3ec67d907b523263224645c9360253b53d555253d81cd16f63fd9a50b","retrievedAt":"2026-11-09T05:32:30.352Z","kind":"publisher-page","preparer":{"model":"server-excerpt-repair/stream-excerpt-repair-v1 (original draft: gpt-6-luna)","runId":"9a80c3cb-1e2a-4c5c-8ca2-77d496d1086f"},"reviewer":{"model":"gpt-6-luna","runId":"5e87c852-701c-482f-8d7d-c2c72a929f7c"}},"audit":{"status":"uncertain","reviewedAt":"2026-11-10T18:50:12.923Z","model":"anthropic/claude-opus-5.5"},"sourceAttribution":{"authors":[{"name":"Hatamizadeh, Ali","url":null,"kind":"unknown"},{"name":"Choi, Yejin","url":null,"kind":"unknown"},{"name":"Kautz, Jan","url":null,"kind":"unknown"}],"authorRole":"article-author","hosts":[],"status":"stated","authorRoster":{"truncated":false,"omitted":0,"assertionGroupsOmitted":0},"participation":[{"actor":{"name":"Hatamizadeh, Ali","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Hatamizadeh, Ali","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2605.22791","locator":"head/meta[name=author|citation_author]"},"assertionId":"17148dfb00b2fe13efa2a41f48f0abfc7d4afa9eec674ef2c0543905137ea21d:participation:0","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-11T11:54:06.837Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"17148dfb00b2fe13efa2a41f48f0abfc7d4afa9eec674ef2c0543905137ea21d","sourceUrl":"https://arxiv.org/abs/2605.22791","contentSha256":"996747306c382c6e5be1a5a0e6f409d843d20c83cbc6de7d0505f780ab38d042","textHash":"f3f6dd0e01bc93c2b075466ec451e7ca79310c41dda70b6388f37c48777fca9b","observedAt":"2026-11-11T11:54:06.837Z"}},{"actor":{"name":"Choi, Yejin","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Choi, Yejin","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2605.22791","locator":"head/meta[name=author|citation_author]"},"assertionId":"17148dfb00b2fe13efa2a41f48f0abfc7d4afa9eec674ef2c0543905137ea21d:participation:1","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-11T11:54:06.837Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"17148dfb00b2fe13efa2a41f48f0abfc7d4afa9eec674ef2c0543905137ea21d","sourceUrl":"https://arxiv.org/abs/2605.22791","contentSha256":"996747306c382c6e5be1a5a0e6f409d843d20c83cbc6de7d0505f780ab38d042","textHash":"f3f6dd0e01bc93c2b075466ec451e7ca79310c41dda70b6388f37c48777fca9b","observedAt":"2026-11-11T11:54:06.837Z"}},{"actor":{"name":"Kautz, Jan","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Kautz, Jan","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2605.22791","locator":"head/meta[name=author|citation_author]"},"assertionId":"17148dfb00b2fe13efa2a41f48f0abfc7d4afa9eec674ef2c0543905137ea21d:participation:2","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-11T11:54:06.837Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"17148dfb00b2fe13efa2a41f48f0abfc7d4afa9eec674ef2c0543905137ea21d","sourceUrl":"https://arxiv.org/abs/2605.22791","contentSha256":"996747306c382c6e5be1a5a0e6f409d843d20c83cbc6de7d0505f780ab38d042","textHash":"f3f6dd0e01bc93c2b075466ec451e7ca79310c41dda70b6388f37c48777fca9b","observedAt":"2026-11-11T11:54:06.837Z"}}],"participationCoverage":{"total":3,"returned":3,"complete":true,"fullAssertions":"source-metadata-endpoint"},"distribution":null,"publisher":null,"sourceUrl":"https://arxiv.org/abs/2605.22791","snapshotId":"17148dfb00b2fe13efa2a41f48f0abfc7d4afa9eec674ef2c0543905137ea21d","contentSha256":"996747306c382c6e5be1a5a0e6f409d843d20c83cbc6de7d0505f780ab38d042","textHash":"f3f6dd0e01bc93c2b075466ec451e7ca79310c41dda70b6388f37c48777fca9b","observedAt":"2026-11-11T11:54:06.837Z","basis":"publisher-page-metadata","sourceLinks":44,"linksTruncated":false,"metadataUrl":"/api/pipeline/stream/b9a32c33bd8c9037a013d311d5b9f3285af49b9fa6600368ed43b33d4b81ea44/source"}},"canonicalUrl":"https://escapemeridian.com/readings/gated-deltanet-2-decoupling-erase-and-write-in-linear-attention-b9a32c33","formats":{"html":"/readings/gated-deltanet-2-decoupling-erase-and-write-in-linear-attention-b9a32c33","markdown":"/readings/gated-deltanet-2-decoupling-erase-and-write-in-linear-attention-b9a32c33.md","json":"/readings/gated-deltanet-2-decoupling-erase-and-write-in-linear-attention-b9a32c33.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-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":"b9a32c33bd8c9037a013d311d5b9f3285af49b9fa6600368ed43b33d4b81ea44","revision":"51a904f3-8e05-40e5-82d3-fd9d2e47be54","revisionKind":"publication-id","title":"Gated DeltaNet-2: Decoupling Erase and Write in Linear Attention","url":"https://escapemeridian.com/readings/gated-deltanet-2-decoupling-erase-and-write-in-linear-attention-b9a32c33"},"source":{"type":"source","id":"https://arxiv.org/abs/2605.22791","revision":"f3f6dd0e01bc93c2b075466ec451e7ca79310c41dda70b6388f37c48777fca9b","revisionKind":"sha256-source-text","title":"Gated DeltaNet-2: Decoupling Erase and Write in Linear Attention","url":"https://arxiv.org/abs/2605.22791"},"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":"b9a32c33bd8c9037a013d311d5b9f3285af49b9fa6600368ed43b33d4b81ea44","title":"Gated DeltaNet-2: Decoupling Erase and Write in Linear Attention","argument":"Ali Hatamizadeh, Yejin Choi and Jan Kautz introduce Gated DeltaNet-2, a linear attention model that separates two operations in its compressed recurrent memory: erasing old content and writing new content. They argue that earlier delta-rule models tied these operations to one scalar gate, limiting how memory could be edited. Their method uses separate channel-wise erase and write gates, while retaining adaptive forgetting and channel-wise decay. The paper also describes a chunkwise WY algorithm and a gate-aware backward pass for parallel training. In experiments with a 1.3B-parameter model trained on 100B FineWeb-Edu tokens, the authors report the strongest overall results among the compared Mamba-2, Gated DeltaNet, KDA and Mamba-3 variants across language modeling, commonsense reasoning and retrieval. They say the advantage is most pronounced on long-context RULER needle-in-a-haystack benchmarks, particularly multi-key retrieval, and report strong results in recurrent and hybrid settings. These findings concern the reported model size, training data and evaluations; they do not establish performance in other settings.","topicIds":["intelligence-and-discovery"],"domain":"science","claimType":"attributed reading","limits":"The reported results are tied to the stated model size, training data, compared variants and evaluations; the evidence does not establish performance beyond those settings. The source gives aggregate claims about benchmark results but does not provide numerical scores in the supplied abstract.","speaker":{"name":"Hatamizadeh, Ali, Choi, Yejin, Kautz, Jan","roleAtSource":"Article author"},"source":{"title":"Gated DeltaNet-2: Decoupling Erase and Write in Linear Attention","publisher":"arXiv cs.AI","format":"Article","url":"https://arxiv.org/abs/2605.22791","publishedAt":"2026-05-21","dateBasis":"Date basis not provided."},"evidence":[{"text":"The paper introduces separate channel-wise erase and write gates to address the scalar tie between erasing old content and committing new content.","excerpt":"the active edit still uses a single scalar gate to control two different things: how much old content to erase on the key side and how much new content to commit on the value side.","url":"https://arxiv.org/abs/2605.22791","relationship":"Source-stated"},{"text":"The authors report the strongest overall results among the compared model variants across three task areas in their stated training setup.","excerpt":"At 1.3B parameters trained on 100B FineWeb-Edu tokens, Gated DeltaNet-2 achieves the strongest overall results among Mamba-2, Gated DeltaNet, KDA, and Mamba-3 variants across language modeling, commonsense reasoning, and retrieval.","url":"https://arxiv.org/abs/2605.22791","relationship":"Source-stated"},{"text":"The authors report the clearest advantage on long-context RULER needle-in-a-haystack benchmarks, including multi-key retrieval.","excerpt":"Its advantage is most pronounced on long-context RULER needle-in-a-haystack benchmarks, where it improves the evaluated multi-key retrieval setting and remains strong in both recurrent and hybrid settings.","url":"https://arxiv.org/abs/2605.22791","relationship":"Source-stated"}],"sourceMedia":{"status":"not_found","sourceUrl":"https://arxiv.org/abs/2605.22791","checkedAt":"2026-11-11T11:54:06.837Z","items":[],"note":"No suitable media was found by the supported page extraction."},"assumptions":[],"openQuestion":{"attribution":"Open question","text":"How would Gated DeltaNet-2 perform on long-context retrieval beyond the evaluated RULER settings?"},"whatWouldChangeView":{"attribution":"Change condition not recorded","text":""},"editorialComparisons":[],"live":{"title":"Gated DeltaNet-2: Decoupling Erase and Write in Linear Attention","summary":"Ali Hatamizadeh, Yejin Choi and Jan Kautz introduce Gated DeltaNet-2, a linear attention model that separates two operations in its compressed recurrent memory: erasing old content and writing new content. They argue that earlier delta-rule models tied these operations to one scalar gate, limiting how memory could be edited. Their method uses separate channel-wise erase and write gates, while retaining adaptive forgetting and channel-wise decay. The paper also describes a chunkwise WY algorithm and a gate-aware backward pass for parallel training. In experiments with a 1.3B-parameter model trained on 100B FineWeb-Edu tokens, the authors report the strongest overall results among the compared Mamba-2, Gated DeltaNet, KDA and Mamba-3 variants across language modeling, commonsense reasoning and retrieval. They say the advantage is most pronounced on long-context RULER needle-in-a-haystack benchmarks, particularly multi-key retrieval, and report strong results in recurrent and hybrid settings. These findings concern the reported model size, training data and evaluations; they do not establish performance in other settings.","question":"How would Gated DeltaNet-2 perform on long-context retrieval beyond the evaluated RULER settings?","claims":[{"text":"The paper introduces separate channel-wise erase and write gates to address the scalar tie between erasing old content and committing new content.","excerpt":"the active edit still uses a single scalar gate to control two different things: how much old content to erase on the key side and how much new content to commit on the value side.","support":"source-stated"},{"text":"The authors report the strongest overall results among the compared model variants across three task areas in their stated training setup.","excerpt":"At 1.3B parameters trained on 100B FineWeb-Edu tokens, Gated DeltaNet-2 achieves the strongest overall results among Mamba-2, Gated DeltaNet, KDA, and Mamba-3 variants across language modeling, commonsense reasoning, and retrieval.","support":"source-stated"},{"text":"The authors report the clearest advantage on long-context RULER needle-in-a-haystack benchmarks, including multi-key retrieval.","excerpt":"Its advantage is most pronounced on long-context RULER needle-in-a-haystack benchmarks, where it improves the evaluated multi-key retrieval setting and remains strong in both recurrent and hybrid settings.","support":"source-stated"}],"limitations":["The reported results are tied to the stated model size, training data, compared variants and evaluations; the evidence does not establish performance beyond those settings.","The source gives aggregate claims about benchmark results but does not provide numerical scores in the supplied abstract."],"mediaId":"93b69d67e028310f9beb9feca15977b198c6f8607f601e6fb23438e6ada8baca","correctionReason":null,"id":"b9a32c33bd8c9037a013d311d5b9f3285af49b9fa6600368ed43b33d4b81ea44","publicationId":"51a904f3-8e05-40e5-82d3-fd9d2e47be54","version":1,"sourceUrl":"https://arxiv.org/abs/2605.22791","sourceTitle":"Gated DeltaNet-2: Decoupling Erase and Write in Linear Attention","publisher":"arXiv cs.AI","publishedAt":null,"originalPublished":{"value":"2026-05-21","precision":"day","basis":"publisher_citation_date"},"firstObservedAt":"2026-11-09T05:04:08.699Z","createdAt":"2026-11-10T18:50:12.923Z","historical":true,"entityIds":["arxiv"],"topicIds":["intelligence-and-discovery"],"media":[{"id":"93b69d67e028310f9beb9feca15977b198c6f8607f601e6fb23438e6ada8baca","kind":"article","url":"https://arxiv.org/abs/2605.22791","sourceUrl":"https://arxiv.org/abs/2605.22791","label":"Read original","provenance":"source_link","verification":{"status":"verified","checkedAt":"2026-11-09T05:32:30.649Z","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/2605.22791","checkedAt":"2026-11-11T11:54:06.837Z","items":[],"note":"No suitable media was found by the supported page extraction."},"editorialDecision":{"outcome":"accept","reasonCodes":["material_development"],"rubricVersion":"stream-admission-v5","mediaId":"93b69d67e028310f9beb9feca15977b198c6f8607f601e6fb23438e6ada8baca","confidence":0.93,"provider":"typesafe","model":"jev-1.13.0","inputHash":"4eedfd150e1d66243dc6f4aaf20f6f097fc64de1ada5ec899664d1211d4ccde3","decidedAt":"2026-11-09T05:32:32.553Z"},"provenance":{"textHash":"f3f6dd0e01bc93c2b075466ec451e7ca79310c41dda70b6388f37c48777fca9b","observationId":"7b1143e3ec67d907b523263224645c9360253b53d555253d81cd16f63fd9a50b","retrievedAt":"2026-11-09T05:32:30.352Z","kind":"publisher-page","preparer":{"model":"server-excerpt-repair/stream-excerpt-repair-v1 (original draft: gpt-6-luna)","runId":"9a80c3cb-1e2a-4c5c-8ca2-77d496d1086f"},"reviewer":{"model":"gpt-6-luna","runId":"5e87c852-701c-482f-8d7d-c2c72a929f7c"}},"audit":{"status":"uncertain","reviewedAt":"2026-11-10T18:50:12.923Z","model":"anthropic/claude-opus-5.5"},"sourceAttribution":{"authors":[{"name":"Hatamizadeh, Ali","url":null,"kind":"unknown"},{"name":"Choi, Yejin","url":null,"kind":"unknown"},{"name":"Kautz, Jan","url":null,"kind":"unknown"}],"authorRole":"article-author","hosts":[],"status":"stated","authorRoster":{"truncated":false,"omitted":0,"assertionGroupsOmitted":0},"participation":[{"actor":{"name":"Hatamizadeh, Ali","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Hatamizadeh, Ali","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2605.22791","locator":"head/meta[name=author|citation_author]"},"assertionId":"17148dfb00b2fe13efa2a41f48f0abfc7d4afa9eec674ef2c0543905137ea21d:participation:0","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-11T11:54:06.837Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"17148dfb00b2fe13efa2a41f48f0abfc7d4afa9eec674ef2c0543905137ea21d","sourceUrl":"https://arxiv.org/abs/2605.22791","contentSha256":"996747306c382c6e5be1a5a0e6f409d843d20c83cbc6de7d0505f780ab38d042","textHash":"f3f6dd0e01bc93c2b075466ec451e7ca79310c41dda70b6388f37c48777fca9b","observedAt":"2026-11-11T11:54:06.837Z"}},{"actor":{"name":"Choi, Yejin","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Choi, Yejin","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2605.22791","locator":"head/meta[name=author|citation_author]"},"assertionId":"17148dfb00b2fe13efa2a41f48f0abfc7d4afa9eec674ef2c0543905137ea21d:participation:1","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-11T11:54:06.837Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"17148dfb00b2fe13efa2a41f48f0abfc7d4afa9eec674ef2c0543905137ea21d","sourceUrl":"https://arxiv.org/abs/2605.22791","contentSha256":"996747306c382c6e5be1a5a0e6f409d843d20c83cbc6de7d0505f780ab38d042","textHash":"f3f6dd0e01bc93c2b075466ec451e7ca79310c41dda70b6388f37c48777fca9b","observedAt":"2026-11-11T11:54:06.837Z"}},{"actor":{"name":"Kautz, Jan","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Kautz, Jan","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2605.22791","locator":"head/meta[name=author|citation_author]"},"assertionId":"17148dfb00b2fe13efa2a41f48f0abfc7d4afa9eec674ef2c0543905137ea21d:participation:2","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-11T11:54:06.837Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"17148dfb00b2fe13efa2a41f48f0abfc7d4afa9eec674ef2c0543905137ea21d","sourceUrl":"https://arxiv.org/abs/2605.22791","contentSha256":"996747306c382c6e5be1a5a0e6f409d843d20c83cbc6de7d0505f780ab38d042","textHash":"f3f6dd0e01bc93c2b075466ec451e7ca79310c41dda70b6388f37c48777fca9b","observedAt":"2026-11-11T11:54:06.837Z"}}],"participationCoverage":{"total":3,"returned":3,"complete":true,"fullAssertions":"source-metadata-endpoint"},"distribution":null,"publisher":null,"sourceUrl":"https://arxiv.org/abs/2605.22791","snapshotId":"17148dfb00b2fe13efa2a41f48f0abfc7d4afa9eec674ef2c0543905137ea21d","contentSha256":"996747306c382c6e5be1a5a0e6f409d843d20c83cbc6de7d0505f780ab38d042","textHash":"f3f6dd0e01bc93c2b075466ec451e7ca79310c41dda70b6388f37c48777fca9b","observedAt":"2026-11-11T11:54:06.837Z","basis":"publisher-page-metadata","sourceLinks":44,"linksTruncated":false,"metadataUrl":"/api/pipeline/stream/b9a32c33bd8c9037a013d311d5b9f3285af49b9fa6600368ed43b33d4b81ea44/source"}}}],"audits":[{"publicationId":"51a904f3-8e05-40e5-82d3-fd9d2e47be54","verdict":"uncertain","findings":[{"dimension":"attribution","status":"pass","explanation":"The draft correctly names authors Ali Hatamizadeh, Yejin Choi and Jan Kautz. It frames the method and results as the authors' claims and reported results rather than established fact.","excerpt":"Authors:Ali Hatamizadeh, Yejin Choi, Jan Kautz"},{"dimension":"summary","status":"uncertain","explanation":"The summary is faithful but hard for a non-specialist to follow. Central terms are not explained: linear attention, scalar gate, channel-wise, chunkwise WY algorithm, backward pass, RULER needle-in-a-haystack, and the Mamba and KDA baselines. It also omits why linear attention matters: the source says a fixed-size state replaces softmax attention's unbounded cache, giving linear-time mixing and constant-memory decoding. Smallest repair: add one plain-language sentence on that efficiency motive and why editing compressed memory is hard. Keep the technical terms exact, with tooltips. Title and question are acceptable.","excerpt":"Linear attention replaces the unbounded cache of softmax attention with a fixed-size recurrent state, reducing sequence mixing to linear time and decoding to constant memory."},{"dimension":"date","status":"pass","explanation":"evidence.originalPublished is 2026-05-21, which matches the v1 submission date. The 6 Nov 2026 revision and the November feed date are correctly not used as the original publication date.","excerpt":"[Submitted on 21 May 2026 (v1), last revised 6 Nov 2026 (this version, v2)]"},{"dimension":"support","status":"pass","explanation":"The claims match the abstract: separate erase and write gates, strongest overall results among the named baselines at 1.3B parameters and 100B tokens, and the most pronounced advantage on RULER, including multi-key retrieval. The limitations properly confine the results to the evaluated setup and note that the abstract gives no numerical scores. The question follows from the evidence without leading.","excerpt":"Gated Delta Rule-2 separates these roles with a channel-wise erase gate b_t and a channel-wise write gate w_t"},{"dimension":"media","status":"pass","explanation":"The selected media is the verified arXiv abstract page for this paper. It is the direct original source and the most relevant access point.","excerpt":null}],"createdAt":"2026-11-10T18:50:12.923Z","model":"anthropic/claude-opus-5.5"}],"correctionUrl":"https://escapemeridian.com/corrections/"}