{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2023:B5ZC7VKUFDFZGGNMQH6EVIT22O","short_pith_number":"pith:B5ZC7VKU","schema_version":"1.0","canonical_sha256":"0f722fd55428cb9319ac81fc4aa27ad39b200b4ff30cb377b84bd17c34856c29","source":{"kind":"arxiv","id":"2306.00946","version":2},"attestation_state":"computed","paper":{"title":"Exposing Attention Glitches with Flip-Flop Language Modeling","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.CL"],"primary_cat":"cs.LG","authors_text":"Akshay Krishnamurthy, Bingbin Liu, Cyril Zhang, Jordan T. Ash, Surbhi Goel","submitted_at":"2023-06-01T17:44:35Z","abstract_excerpt":"Why do large language models sometimes output factual inaccuracies and exhibit erroneous reasoning? The brittleness of these models, particularly when executing long chains of reasoning, currently seems to be an inevitable price to pay for their advanced capabilities of coherently synthesizing knowledge, pragmatics, and abstract thought. Towards making sense of this fundamentally unsolved problem, this work identifies and analyzes the phenomenon of attention glitches, in which the Transformer architecture's inductive biases intermittently fail to capture robust reasoning. To isolate the issue,"},"verification_status":{"content_addressed":true,"pith_receipt":true,"author_attested":false,"weak_author_claims":0,"strong_author_claims":0,"externally_anchored":false,"storage_verified":false,"citation_signatures":0,"replication_records":0,"graph_snapshot":true,"references_resolved":false,"formal_links_present":false},"canonical_record":{"source":{"id":"2306.00946","kind":"arxiv","version":2},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2023-06-01T17:44:35Z","cross_cats_sorted":["cs.CL"],"title_canon_sha256":"d90983e85dd94d44746ac2aed4fb973a868e7eb0103b0f841761ec1eadc3e8b9","abstract_canon_sha256":"5b233fa65283689e120acf160b3a3b114f59e2a23dab604bb4a1d3f45ecab735"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T07:06:29.049613Z","signature_b64":"GgXXHwAeKWHJrhC4LVlsMoByNSdWhgp09V+XV9SBVpQ1W1YopRiGYREah3Kse4UUItXp+nIG8I6QsSY68PuODg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"0f722fd55428cb9319ac81fc4aa27ad39b200b4ff30cb377b84bd17c34856c29","last_reissued_at":"2026-07-05T07:06:29.049135Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T07:06:29.049135Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Exposing Attention Glitches with Flip-Flop Language Modeling","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.CL"],"primary_cat":"cs.LG","authors_text":"Akshay Krishnamurthy, Bingbin Liu, Cyril Zhang, Jordan T. Ash, Surbhi Goel","submitted_at":"2023-06-01T17:44:35Z","abstract_excerpt":"Why do large language models sometimes output factual inaccuracies and exhibit erroneous reasoning? The brittleness of these models, particularly when executing long chains of reasoning, currently seems to be an inevitable price to pay for their advanced capabilities of coherently synthesizing knowledge, pragmatics, and abstract thought. Towards making sense of this fundamentally unsolved problem, this work identifies and analyzes the phenomenon of attention glitches, in which the Transformer architecture's inductive biases intermittently fail to capture robust reasoning. To isolate the issue,"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2306.00946","kind":"arxiv","version":2},"verdict":{"id":null,"model_set":{},"created_at":null,"strongest_claim":"","one_line_summary":"","pipeline_version":null,"weakest_assumption":"","pith_extraction_headline":""},"integrity":{"clean":true,"summary":{"advisory":0,"critical":0,"by_detector":{},"informational":0},"endpoint":"/pith/2306.00946/integrity.json","findings":[],"available":true,"detectors_run":[],"snapshot_sha256":"c28c3603d3b5d939e8dc4c7e95fa8dfce3d595e45f758748cecf8e644a296938"},"references":{"count":0,"sample":[],"resolved_work":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57","internal_anchors":0},"formal_canon":{"evidence_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"},"aliases":[{"alias_kind":"arxiv","alias_value":"2306.00946","created_at":"2026-07-05T07:06:29.049193+00:00"},{"alias_kind":"arxiv_version","alias_value":"2306.00946v2","created_at":"2026-07-05T07:06:29.049193+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2306.00946","created_at":"2026-07-05T07:06:29.049193+00:00"},{"alias_kind":"pith_short_12","alias_value":"B5ZC7VKUFDFZ","created_at":"2026-07-05T07:06:29.049193+00:00"},{"alias_kind":"pith_short_16","alias_value":"B5ZC7VKUFDFZGGNM","created_at":"2026-07-05T07:06:29.049193+00:00"},{"alias_kind":"pith_short_8","alias_value":"B5ZC7VKU","created_at":"2026-07-05T07:06:29.049193+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":2,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2311.05232","citing_title":"A Survey on Hallucination in Large Language Models: Principles, Taxonomy, Challenges, and Open Questions","ref_index":190,"is_internal_anchor":false},{"citing_arxiv_id":"2604.05923","citing_title":"The UNDO Flip-Flop: A Controlled Probe for Reversible Semantic State Management in State Space Model","ref_index":10,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/B5ZC7VKUFDFZGGNMQH6EVIT22O","json":"https://pith.science/pith/B5ZC7VKUFDFZGGNMQH6EVIT22O.json","graph_json":"https://pith.science/api/pith-number/B5ZC7VKUFDFZGGNMQH6EVIT22O/graph.json","events_json":"https://pith.science/api/pith-number/B5ZC7VKUFDFZGGNMQH6EVIT22O/events.json","paper":"https://pith.science/paper/B5ZC7VKU"},"agent_actions":{"view_html":"https://pith.science/pith/B5ZC7VKUFDFZGGNMQH6EVIT22O","download_json":"https://pith.science/pith/B5ZC7VKUFDFZGGNMQH6EVIT22O.json","view_paper":"https://pith.science/paper/B5ZC7VKU","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2306.00946&json=true","fetch_graph":"https://pith.science/api/pith-number/B5ZC7VKUFDFZGGNMQH6EVIT22O/graph.json","fetch_events":"https://pith.science/api/pith-number/B5ZC7VKUFDFZGGNMQH6EVIT22O/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/B5ZC7VKUFDFZGGNMQH6EVIT22O/action/timestamp_anchor","attest_storage":"https://pith.science/pith/B5ZC7VKUFDFZGGNMQH6EVIT22O/action/storage_attestation","attest_author":"https://pith.science/pith/B5ZC7VKUFDFZGGNMQH6EVIT22O/action/author_attestation","sign_citation":"https://pith.science/pith/B5ZC7VKUFDFZGGNMQH6EVIT22O/action/citation_signature","submit_replication":"https://pith.science/pith/B5ZC7VKUFDFZGGNMQH6EVIT22O/action/replication_record"}},"created_at":"2026-07-05T07:06:29.049193+00:00","updated_at":"2026-07-05T07:06:29.049193+00:00"}