{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:66QXZ4FJYTGD4ZNQASST5IPCR7","short_pith_number":"pith:66QXZ4FJ","schema_version":"1.0","canonical_sha256":"f7a17cf0a9c4cc3e65b004a53ea1e28fecee601fca03d6d71a7428372971a6cb","source":{"kind":"arxiv","id":"2503.08908","version":1},"attestation_state":"computed","paper":{"title":"Interpreting the Repeated Token Phenomenon in Large Language Models","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI","cs.CL","cs.CR"],"primary_cat":"cs.LG","authors_text":"Federico Barbero, Ilia Shumailov, Itay Yona, Jamie Hayes, Yossi Gandelsman","submitted_at":"2025-03-11T21:40:58Z","abstract_excerpt":"Large Language Models (LLMs), despite their impressive capabilities, often fail to accurately repeat a single word when prompted to, and instead output unrelated text. This unexplained failure mode represents a vulnerability, allowing even end-users to diverge models away from their intended behavior. We aim to explain the causes for this phenomenon and link it to the concept of ``attention sinks'', an emergent LLM behavior crucial for fluency, in which the initial token receives disproportionately high attention scores. Our investigation identifies the neural circuit responsible for attention"},"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":"2503.08908","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2025-03-11T21:40:58Z","cross_cats_sorted":["cs.AI","cs.CL","cs.CR"],"title_canon_sha256":"3673431520ce2132b67d497789061d67ea330a1426670ea618def70c0a55f73b","abstract_canon_sha256":"968768bbe6373d20e73b80e66504078875a98df2e0a4661c8af717f8317a524d"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T10:29:43.496782Z","signature_b64":"65emIqXhGYTbhQe2Q+sBgK6rAU97IjaQwn3AkzERB0XUg58pSVCT80r1zzxH+p+tLu26trIVaD5nlFX1OR48CQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"f7a17cf0a9c4cc3e65b004a53ea1e28fecee601fca03d6d71a7428372971a6cb","last_reissued_at":"2026-07-05T10:29:43.495974Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T10:29:43.495974Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Interpreting the Repeated Token Phenomenon in Large Language Models","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI","cs.CL","cs.CR"],"primary_cat":"cs.LG","authors_text":"Federico Barbero, Ilia Shumailov, Itay Yona, Jamie Hayes, Yossi Gandelsman","submitted_at":"2025-03-11T21:40:58Z","abstract_excerpt":"Large Language Models (LLMs), despite their impressive capabilities, often fail to accurately repeat a single word when prompted to, and instead output unrelated text. This unexplained failure mode represents a vulnerability, allowing even end-users to diverge models away from their intended behavior. We aim to explain the causes for this phenomenon and link it to the concept of ``attention sinks'', an emergent LLM behavior crucial for fluency, in which the initial token receives disproportionately high attention scores. Our investigation identifies the neural circuit responsible for attention"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2503.08908","kind":"arxiv","version":1},"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/2503.08908/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":"2503.08908","created_at":"2026-07-05T10:29:43.496071+00:00"},{"alias_kind":"arxiv_version","alias_value":"2503.08908v1","created_at":"2026-07-05T10:29:43.496071+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2503.08908","created_at":"2026-07-05T10:29:43.496071+00:00"},{"alias_kind":"pith_short_12","alias_value":"66QXZ4FJYTGD","created_at":"2026-07-05T10:29:43.496071+00:00"},{"alias_kind":"pith_short_16","alias_value":"66QXZ4FJYTGD4ZNQ","created_at":"2026-07-05T10:29:43.496071+00:00"},{"alias_kind":"pith_short_8","alias_value":"66QXZ4FJ","created_at":"2026-07-05T10:29:43.496071+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":5,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.07604","citing_title":"Contribution Weights: A Geometrical Analysis of Self-Attention Transformers","ref_index":85,"is_internal_anchor":false},{"citing_arxiv_id":"2605.16147","citing_title":"Registers Matter for Pixel-Space Diffusion Transformers","ref_index":50,"is_internal_anchor":false},{"citing_arxiv_id":"2505.06708","citing_title":"Gated Attention for Large Language Models: Non-linearity, Sparsity, and Attention-Sink-Free","ref_index":31,"is_internal_anchor":false},{"citing_arxiv_id":"2605.06611","citing_title":"The Structural Origin of Attention Sink: Variance Discrepancy, Super Neurons, and Dimension Disparity","ref_index":31,"is_internal_anchor":false},{"citing_arxiv_id":"2605.05206","citing_title":"Taming Outlier Tokens in Diffusion Transformers","ref_index":36,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/66QXZ4FJYTGD4ZNQASST5IPCR7","json":"https://pith.science/pith/66QXZ4FJYTGD4ZNQASST5IPCR7.json","graph_json":"https://pith.science/api/pith-number/66QXZ4FJYTGD4ZNQASST5IPCR7/graph.json","events_json":"https://pith.science/api/pith-number/66QXZ4FJYTGD4ZNQASST5IPCR7/events.json","paper":"https://pith.science/paper/66QXZ4FJ"},"agent_actions":{"view_html":"https://pith.science/pith/66QXZ4FJYTGD4ZNQASST5IPCR7","download_json":"https://pith.science/pith/66QXZ4FJYTGD4ZNQASST5IPCR7.json","view_paper":"https://pith.science/paper/66QXZ4FJ","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2503.08908&json=true","fetch_graph":"https://pith.science/api/pith-number/66QXZ4FJYTGD4ZNQASST5IPCR7/graph.json","fetch_events":"https://pith.science/api/pith-number/66QXZ4FJYTGD4ZNQASST5IPCR7/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/66QXZ4FJYTGD4ZNQASST5IPCR7/action/timestamp_anchor","attest_storage":"https://pith.science/pith/66QXZ4FJYTGD4ZNQASST5IPCR7/action/storage_attestation","attest_author":"https://pith.science/pith/66QXZ4FJYTGD4ZNQASST5IPCR7/action/author_attestation","sign_citation":"https://pith.science/pith/66QXZ4FJYTGD4ZNQASST5IPCR7/action/citation_signature","submit_replication":"https://pith.science/pith/66QXZ4FJYTGD4ZNQASST5IPCR7/action/replication_record"}},"created_at":"2026-07-05T10:29:43.496071+00:00","updated_at":"2026-07-05T10:29:43.496071+00:00"}