{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2023:YSN63TOHXNTREP47WS74OONWG2","short_pith_number":"pith:YSN63TOH","schema_version":"1.0","canonical_sha256":"c49bedcdc7bb67123f9fb4bfc739b636bd9cfaf70aca56b3ac49c08911b07c72","source":{"kind":"arxiv","id":"2311.09615","version":2},"attestation_state":"computed","paper":{"title":"On Retrieval Augmentation and the Limitations of Language Model Training","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Dani Yogatama, Isabelle Lee, Joshua Robinson, Ollie Liu, Ting-Rui Chiang, Xinyan Velocity Yu","submitted_at":"2023-11-16T06:59:54Z","abstract_excerpt":"Augmenting a language model (LM) with $k$-nearest neighbors ($k$NN) retrieval on its training data alone can decrease its perplexity, though the underlying reasons for this remain elusive. In this work, we rule out one previously posited possibility -- the \"softmax bottleneck.\" We then create a new dataset to evaluate LM generalization ability in the setting where training data contains additional information that is not causally relevant. This task is challenging even for GPT-3.5 Turbo. We show that, for both GPT-2 and Mistral 7B, $k$NN retrieval augmentation consistently improves performance"},"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":"2311.09615","kind":"arxiv","version":2},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CL","submitted_at":"2023-11-16T06:59:54Z","cross_cats_sorted":[],"title_canon_sha256":"26c1938f390893b2c8f4bea29b4b8fb5dfc2100b24d4cee8fb69c8490698b1ba","abstract_canon_sha256":"c536adda80fce3caab0862d99fc20ac79d35822c356bf6976dd0c6978d1010d0"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T08:03:08.686119Z","signature_b64":"fQIsWQ/SA1M9aYrty9BIir5x99UU6HqYzkQMx+V5ttgaEB8Is7kopvcw5dgQXeqoQ//r5CIDjx/TnRD6Z5v4BA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"c49bedcdc7bb67123f9fb4bfc739b636bd9cfaf70aca56b3ac49c08911b07c72","last_reissued_at":"2026-07-05T08:03:08.685609Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T08:03:08.685609Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"On Retrieval Augmentation and the Limitations of Language Model Training","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Dani Yogatama, Isabelle Lee, Joshua Robinson, Ollie Liu, Ting-Rui Chiang, Xinyan Velocity Yu","submitted_at":"2023-11-16T06:59:54Z","abstract_excerpt":"Augmenting a language model (LM) with $k$-nearest neighbors ($k$NN) retrieval on its training data alone can decrease its perplexity, though the underlying reasons for this remain elusive. In this work, we rule out one previously posited possibility -- the \"softmax bottleneck.\" We then create a new dataset to evaluate LM generalization ability in the setting where training data contains additional information that is not causally relevant. This task is challenging even for GPT-3.5 Turbo. We show that, for both GPT-2 and Mistral 7B, $k$NN retrieval augmentation consistently improves performance"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2311.09615","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/2311.09615/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":"2311.09615","created_at":"2026-07-05T08:03:08.685669+00:00"},{"alias_kind":"arxiv_version","alias_value":"2311.09615v2","created_at":"2026-07-05T08:03:08.685669+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2311.09615","created_at":"2026-07-05T08:03:08.685669+00:00"},{"alias_kind":"pith_short_12","alias_value":"YSN63TOHXNTR","created_at":"2026-07-05T08:03:08.685669+00:00"},{"alias_kind":"pith_short_16","alias_value":"YSN63TOHXNTREP47","created_at":"2026-07-05T08:03:08.685669+00:00"},{"alias_kind":"pith_short_8","alias_value":"YSN63TOH","created_at":"2026-07-05T08:03:08.685669+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":0,"internal_anchor_count":0,"sample":[]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/YSN63TOHXNTREP47WS74OONWG2","json":"https://pith.science/pith/YSN63TOHXNTREP47WS74OONWG2.json","graph_json":"https://pith.science/api/pith-number/YSN63TOHXNTREP47WS74OONWG2/graph.json","events_json":"https://pith.science/api/pith-number/YSN63TOHXNTREP47WS74OONWG2/events.json","paper":"https://pith.science/paper/YSN63TOH"},"agent_actions":{"view_html":"https://pith.science/pith/YSN63TOHXNTREP47WS74OONWG2","download_json":"https://pith.science/pith/YSN63TOHXNTREP47WS74OONWG2.json","view_paper":"https://pith.science/paper/YSN63TOH","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2311.09615&json=true","fetch_graph":"https://pith.science/api/pith-number/YSN63TOHXNTREP47WS74OONWG2/graph.json","fetch_events":"https://pith.science/api/pith-number/YSN63TOHXNTREP47WS74OONWG2/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/YSN63TOHXNTREP47WS74OONWG2/action/timestamp_anchor","attest_storage":"https://pith.science/pith/YSN63TOHXNTREP47WS74OONWG2/action/storage_attestation","attest_author":"https://pith.science/pith/YSN63TOHXNTREP47WS74OONWG2/action/author_attestation","sign_citation":"https://pith.science/pith/YSN63TOHXNTREP47WS74OONWG2/action/citation_signature","submit_replication":"https://pith.science/pith/YSN63TOHXNTREP47WS74OONWG2/action/replication_record"}},"created_at":"2026-07-05T08:03:08.685669+00:00","updated_at":"2026-07-05T08:03:08.685669+00:00"}