{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2024:ATXHXYAHXFA66PSCTLANT76245","short_pith_number":"pith:ATXHXYAH","schema_version":"1.0","canonical_sha256":"04ee7be007b941ef3e429ac0d9ffdae7714f7418ea2c2ce8d07fe78dd0a48833","source":{"kind":"arxiv","id":"2410.02284","version":1},"attestation_state":"computed","paper":{"title":"Correlation and Navigation in the Vocabulary Key Representation Space of Language Models","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Chenyang An, Jingbo Shang, Letian Peng","submitted_at":"2024-10-03T08:07:55Z","abstract_excerpt":"Language model (LM) decoding is based on the next-token prediction (NTP) probability distribution. For neural LMs (e.g., Transformer-based), NTP distribution is essentially a softmax-regularized dot product between an encoded input context (query) and fixed vocabulary representations (keys). In this paper, we study the effect of the key distribution on the NTP distribution, with a focus on whether the similarity between keys will trigger spurious correlations in NTP. Through knowledge-probing tasks, we show that in the NTP distribution, the few top-ranked tokens are typically accurate. However"},"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":"2410.02284","kind":"arxiv","version":1},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CL","submitted_at":"2024-10-03T08:07:55Z","cross_cats_sorted":[],"title_canon_sha256":"1c0e5116fffc2ead3265b347c0e2557caa03a3e0c955a71a7d4e95cf3bf74172","abstract_canon_sha256":"b7cc114e49423d7e1b80b101cd1a8962c2be19f001fec6a7419c122c2a77a863"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T09:15:18.786728Z","signature_b64":"pQsF2f8NenEJiG2mkCUQo+KnUvo8tLWt2hIBDA0kfrS0D6xNMXpb8cvLW6JM32FKHcnQzhWYl418UApV8eNjDQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"04ee7be007b941ef3e429ac0d9ffdae7714f7418ea2c2ce8d07fe78dd0a48833","last_reissued_at":"2026-07-05T09:15:18.786273Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T09:15:18.786273Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Correlation and Navigation in the Vocabulary Key Representation Space of Language Models","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Chenyang An, Jingbo Shang, Letian Peng","submitted_at":"2024-10-03T08:07:55Z","abstract_excerpt":"Language model (LM) decoding is based on the next-token prediction (NTP) probability distribution. For neural LMs (e.g., Transformer-based), NTP distribution is essentially a softmax-regularized dot product between an encoded input context (query) and fixed vocabulary representations (keys). In this paper, we study the effect of the key distribution on the NTP distribution, with a focus on whether the similarity between keys will trigger spurious correlations in NTP. Through knowledge-probing tasks, we show that in the NTP distribution, the few top-ranked tokens are typically accurate. However"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2410.02284","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/2410.02284/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":"2410.02284","created_at":"2026-07-05T09:15:18.786335+00:00"},{"alias_kind":"arxiv_version","alias_value":"2410.02284v1","created_at":"2026-07-05T09:15:18.786335+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2410.02284","created_at":"2026-07-05T09:15:18.786335+00:00"},{"alias_kind":"pith_short_12","alias_value":"ATXHXYAHXFA6","created_at":"2026-07-05T09:15:18.786335+00:00"},{"alias_kind":"pith_short_16","alias_value":"ATXHXYAHXFA66PSC","created_at":"2026-07-05T09:15:18.786335+00:00"},{"alias_kind":"pith_short_8","alias_value":"ATXHXYAH","created_at":"2026-07-05T09:15:18.786335+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":1,"internal_anchor_count":1,"sample":[{"citing_arxiv_id":"2502.04520","citing_title":"Linear Correlation in LM's Compositional Generalization and Hallucination","ref_index":28,"is_internal_anchor":true}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/ATXHXYAHXFA66PSCTLANT76245","json":"https://pith.science/pith/ATXHXYAHXFA66PSCTLANT76245.json","graph_json":"https://pith.science/api/pith-number/ATXHXYAHXFA66PSCTLANT76245/graph.json","events_json":"https://pith.science/api/pith-number/ATXHXYAHXFA66PSCTLANT76245/events.json","paper":"https://pith.science/paper/ATXHXYAH"},"agent_actions":{"view_html":"https://pith.science/pith/ATXHXYAHXFA66PSCTLANT76245","download_json":"https://pith.science/pith/ATXHXYAHXFA66PSCTLANT76245.json","view_paper":"https://pith.science/paper/ATXHXYAH","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2410.02284&json=true","fetch_graph":"https://pith.science/api/pith-number/ATXHXYAHXFA66PSCTLANT76245/graph.json","fetch_events":"https://pith.science/api/pith-number/ATXHXYAHXFA66PSCTLANT76245/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/ATXHXYAHXFA66PSCTLANT76245/action/timestamp_anchor","attest_storage":"https://pith.science/pith/ATXHXYAHXFA66PSCTLANT76245/action/storage_attestation","attest_author":"https://pith.science/pith/ATXHXYAHXFA66PSCTLANT76245/action/author_attestation","sign_citation":"https://pith.science/pith/ATXHXYAHXFA66PSCTLANT76245/action/citation_signature","submit_replication":"https://pith.science/pith/ATXHXYAHXFA66PSCTLANT76245/action/replication_record"}},"created_at":"2026-07-05T09:15:18.786335+00:00","updated_at":"2026-07-05T09:15:18.786335+00:00"}