{"state_type":"pith_open_graph_state","state_version":"1.0","pith_number":"pith:2026:DRCNPBOL4LFP4KPSBUUH2GHM3S","merge_version":"pith-open-graph-merge-v1","event_count":2,"valid_event_count":2,"invalid_event_count":0,"equivocation_count":0,"current":{"canonical_record":{"metadata":{"abstract_canon_sha256":"6c8c5dbb6f4bc4438f88cdee9f6f6be59c17805b60a5aa72fe5788b4dc225d20","cross_cats_sorted":["cs.CL"],"license":"http://creativecommons.org/licenses/by-sa/4.0/","primary_cat":"cs.AI","submitted_at":"2026-05-13T02:58:18Z","title_canon_sha256":"2bbf3223d3da58d082e2aedb69c8f93e0a5c66f6ef707e03a30352b81963495e"},"schema_version":"1.0","source":{"id":"2605.12922","kind":"arxiv","version":1}},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2605.12922","created_at":"2026-05-18T03:09:10Z"},{"alias_kind":"arxiv_version","alias_value":"2605.12922v1","created_at":"2026-05-18T03:09:10Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2605.12922","created_at":"2026-05-18T03:09:10Z"},{"alias_kind":"pith_short_12","alias_value":"DRCNPBOL4LFP","created_at":"2026-05-18T12:33:37Z"},{"alias_kind":"pith_short_16","alias_value":"DRCNPBOL4LFP4KPS","created_at":"2026-05-18T12:33:37Z"},{"alias_kind":"pith_short_8","alias_value":"DRCNPBOL","created_at":"2026-05-18T12:33:37Z"}],"graph_snapshots":[{"event_id":"sha256:7f05a9434822ff19a66e7b0262583e357791c4d11f94edec1b80c8567c978a4f","target":"graph","created_at":"2026-05-18T03:09:10Z","signer":{"key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signer_id":"pith.science","signer_type":"pith_registry"},"payload":{"graph_snapshot":{"author_claims":{"count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57","strong_count":0},"builder_version":"pith-number-builder-2026-05-17-v1","claims":{"count":4,"items":[{"attestation":"unclaimed","claim_id":"C1","kind":"strongest_claim","source":"verdict.strongest_claim","status":"machine_extracted","text":"Across architectures and model scales, the gap between attention loss and residual decodability predicts whether goal-conditioned behavior survives channel closure."},{"attestation":"unclaimed","claim_id":"C2","kind":"weakest_assumption","source":"verdict.weakest_assumption","status":"machine_extracted","text":"That the linear probes on residual streams recover causally relevant goal information rather than spurious correlations, and that the sliding-window ablations isolate the attention channel without confounding other mechanisms."},{"attestation":"unclaimed","claim_id":"C3","kind":"one_line_summary","source":"verdict.one_line_summary","status":"machine_extracted","text":"Attention to goal tokens declines in multi-turn LLM interactions while residual representations often retain decodable goal information, and the gap between these predicts whether goal-conditioned behavior survives."},{"attestation":"unclaimed","claim_id":"C4","kind":"headline","source":"verdict.pith_extraction.headline","status":"machine_extracted","text":"LLMs lose track of instructions in multi-turn chats when attention to goal tokens fades, though residual streams may still encode the needed information."}],"snapshot_sha256":"66c60cca120bc073a4b1d313963698529129e4d39ab4a99e19780123d969800d"},"formal_canon":{"evidence_count":2,"snapshot_sha256":"c864106dc6b25539c1110e1faf9125625aec269b7b8ef15cc8dbc800ea61e563"},"paper":{"abstract_excerpt":"Large language models can follow complex instructions in a single turn, yet over long multi-turn interactions they often lose the thread of instructions, persona, and rules. This degradation has been measured behaviorally but not mechanistically explained. We propose a channel-transition account: goal-defining tokens become less accessible through attention, while goal-related information may persist in residual representations. We introduce the Goal Accessibility Ratio (GAR), measuring attention from generated tokens to task-defining goal tokens, and combine it with sliding-window ablations a","authors_text":"Dilek Hakkani-T\\\"ur, Joseph Hsieh, Seunghyun Yoon, Trung Bui, Vardhan Dongre, Viet Dac Lai","cross_cats":["cs.CL"],"headline":"LLMs lose track of instructions in multi-turn chats when attention to goal tokens fades, though residual streams may still encode the needed information.","license":"http://creativecommons.org/licenses/by-sa/4.0/","primary_cat":"cs.AI","submitted_at":"2026-05-13T02:58:18Z","title":"When Attention Closes: How LLMs Lose the Thread in Multi-Turn Interaction"},"references":{"count":39,"internal_anchors":17,"resolved_work":39,"sample":[{"cited_arxiv_id":"2505.06120","doi":"","is_internal_anchor":true,"ref_index":1,"title":"LLMs Get Lost In Multi-Turn Conversation","work_id":"01fa2ab4-4b40-4e8a-9487-2de7db19cfe2","year":null},{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":2,"title":"Multi-if: Benchmarking llms on multi-turn and multilingual instructions following","work_id":"2e46fb28-13e3-4c54-88b5-a2d47f2a064b","year":null},{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":3,"title":"Transactions of the association for computational linguistics , volume=","work_id":"35c18954-8a0c-4935-b39c-5981194920ba","year":null},{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":4,"title":"Roformer: Enhanced transformer with rotary position embedding , author=. Neurocomputing , volume=. 2024 , publisher=","work_id":"0cfd4ae7-837c-4c4f-bf6d-afc10f5a8ed1","year":2024},{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":5,"title":"Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) , pages=","work_id":"756542c0-c2f6-4557-9f1f-50c1be6099a8","year":null}],"snapshot_sha256":"e9bd9f0947179197d22903ff1caab17d502cfdd23f9c82b831f5c672746b816f"},"source":{"id":"2605.12922","kind":"arxiv","version":1},"verdict":{"created_at":"2026-05-14T20:16:15.218921Z","id":"dbd61569-cdfb-4f4d-81c5-ff1b18d7e154","model_set":{"reader":"grok-4.3"},"one_line_summary":"Attention to goal tokens declines in multi-turn LLM interactions while residual representations often retain decodable goal information, and the gap between these predicts whether goal-conditioned behavior survives.","pipeline_version":"pith-pipeline@v0.9.0","pith_extraction_headline":"LLMs lose track of instructions in multi-turn chats when attention to goal tokens fades, though residual streams may still encode the needed information.","strongest_claim":"Across architectures and model scales, the gap between attention loss and residual decodability predicts whether goal-conditioned behavior survives channel closure.","weakest_assumption":"That the linear probes on residual streams recover causally relevant goal information rather than spurious correlations, and that the sliding-window ablations isolate the attention channel without confounding other mechanisms."}},"verdict_id":"dbd61569-cdfb-4f4d-81c5-ff1b18d7e154"}}],"author_attestations":[],"timestamp_anchors":[],"storage_attestations":[],"citation_signatures":[],"replication_records":[],"corrections":[],"mirror_hints":[],"record_created":{"event_id":"sha256:9c2b26a21a3a3bf774f0419962fc4a1bf675822a94eed26b5e4e12e829d66d5d","target":"record","created_at":"2026-05-18T03:09:10Z","signer":{"key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signer_id":"pith.science","signer_type":"pith_registry"},"payload":{"attestation_state":"computed","canonical_record":{"metadata":{"abstract_canon_sha256":"6c8c5dbb6f4bc4438f88cdee9f6f6be59c17805b60a5aa72fe5788b4dc225d20","cross_cats_sorted":["cs.CL"],"license":"http://creativecommons.org/licenses/by-sa/4.0/","primary_cat":"cs.AI","submitted_at":"2026-05-13T02:58:18Z","title_canon_sha256":"2bbf3223d3da58d082e2aedb69c8f93e0a5c66f6ef707e03a30352b81963495e"},"schema_version":"1.0","source":{"id":"2605.12922","kind":"arxiv","version":1}},"canonical_sha256":"1c44d785cbe2cafe29f20d287d18ecdca6420aaac85dabee8dacae5215f2c33d","receipt":{"algorithm":"ed25519","builder_version":"pith-number-builder-2026-05-17-v1","canonical_sha256":"1c44d785cbe2cafe29f20d287d18ecdca6420aaac85dabee8dacae5215f2c33d","first_computed_at":"2026-05-18T03:09:10.199507Z","key_id":"pith-v1-2026-05","kind":"pith_receipt","last_reissued_at":"2026-05-18T03:09:10.199507Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","receipt_version":"0.3","signature_b64":"PrFZW5XzIApoekeR09R3gm0+GuAhBUrzietI90YTozguSIj0PniLyGpYPPRCMzQGTCLqhOdFEbo3DMofisdsBQ==","signature_status":"signed_v1","signed_at":"2026-05-18T03:09:10.200250Z","signed_message":"canonical_sha256_bytes"},"source_id":"2605.12922","source_kind":"arxiv","source_version":1}}},"equivocations":[],"invalid_events":[],"applied_event_ids":["sha256:9c2b26a21a3a3bf774f0419962fc4a1bf675822a94eed26b5e4e12e829d66d5d","sha256:7f05a9434822ff19a66e7b0262583e357791c4d11f94edec1b80c8567c978a4f"],"state_sha256":"946d267594be888b948265febb9ced8a9b7016646232dde15c78b44ba4149de6"}