{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2024:YPNK2L5VV3WQEEIVNUHCQ5SV7B","short_pith_number":"pith:YPNK2L5V","schema_version":"1.0","canonical_sha256":"c3daad2fb5aeed0211156d0e287655f87244778d30f950f255e5c1aab711130f","source":{"kind":"arxiv","id":"2410.01288","version":2},"attestation_state":"computed","paper":{"title":"Mitigating Copy Bias in In-Context Learning through Neuron Pruning","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.LG"],"primary_cat":"cs.CL","authors_text":"Ameen Ali, Ivan Titov, Lior Wolf","submitted_at":"2024-10-02T07:18:16Z","abstract_excerpt":"Large language models (LLMs) have demonstrated impressive few-shot in-context learning (ICL) abilities. Still, we show that they are sometimes prone to a `copying bias', where they copy answers from provided examples instead of learning the underlying patterns. In this work, we propose a novel and simple method to mitigate such copying bias. First, we create a synthetic task and use the Integrated Gradients method to identify neurons that prioritize copying over generalization. We demonstrate that pruning these neurons consistently improves performance across a diverse set of ICL tasks. We als"},"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":"2410.01288","kind":"arxiv","version":2},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CL","submitted_at":"2024-10-02T07:18:16Z","cross_cats_sorted":["cs.LG"],"title_canon_sha256":"4888c1abce6a601d9b1c667d95134d392688885f762000b80ebdecee1b26135c","abstract_canon_sha256":"3f21d6e27035609b232c35ffc6c606596c7235bde2f8aaeb07379eccf3fc3c17"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T09:15:15.221163Z","signature_b64":"Kq9647JtRQ90AHCOzMa+h8y+riXV0MKMHgGXtTnI0Cqi80IM/UpfkIebB1jD7y6jWA+8EGWVb/XNcKJhCT4qAQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"c3daad2fb5aeed0211156d0e287655f87244778d30f950f255e5c1aab711130f","last_reissued_at":"2026-07-05T09:15:15.220757Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T09:15:15.220757Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Mitigating Copy Bias in In-Context Learning through Neuron Pruning","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.LG"],"primary_cat":"cs.CL","authors_text":"Ameen Ali, Ivan Titov, Lior Wolf","submitted_at":"2024-10-02T07:18:16Z","abstract_excerpt":"Large language models (LLMs) have demonstrated impressive few-shot in-context learning (ICL) abilities. Still, we show that they are sometimes prone to a `copying bias', where they copy answers from provided examples instead of learning the underlying patterns. In this work, we propose a novel and simple method to mitigate such copying bias. First, we create a synthetic task and use the Integrated Gradients method to identify neurons that prioritize copying over generalization. We demonstrate that pruning these neurons consistently improves performance across a diverse set of ICL tasks. We als"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2410.01288","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/2410.01288/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":"2410.01288","created_at":"2026-07-05T09:15:15.220811+00:00"},{"alias_kind":"arxiv_version","alias_value":"2410.01288v2","created_at":"2026-07-05T09:15:15.220811+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2410.01288","created_at":"2026-07-05T09:15:15.220811+00:00"},{"alias_kind":"pith_short_12","alias_value":"YPNK2L5VV3WQ","created_at":"2026-07-05T09:15:15.220811+00:00"},{"alias_kind":"pith_short_16","alias_value":"YPNK2L5VV3WQEEIV","created_at":"2026-07-05T09:15:15.220811+00:00"},{"alias_kind":"pith_short_8","alias_value":"YPNK2L5V","created_at":"2026-07-05T09:15:15.220811+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":1,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.18453","citing_title":"LLM Parameters for Math Across Languages: Shared or Separate?","ref_index":28,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/YPNK2L5VV3WQEEIVNUHCQ5SV7B","json":"https://pith.science/pith/YPNK2L5VV3WQEEIVNUHCQ5SV7B.json","graph_json":"https://pith.science/api/pith-number/YPNK2L5VV3WQEEIVNUHCQ5SV7B/graph.json","events_json":"https://pith.science/api/pith-number/YPNK2L5VV3WQEEIVNUHCQ5SV7B/events.json","paper":"https://pith.science/paper/YPNK2L5V"},"agent_actions":{"view_html":"https://pith.science/pith/YPNK2L5VV3WQEEIVNUHCQ5SV7B","download_json":"https://pith.science/pith/YPNK2L5VV3WQEEIVNUHCQ5SV7B.json","view_paper":"https://pith.science/paper/YPNK2L5V","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2410.01288&json=true","fetch_graph":"https://pith.science/api/pith-number/YPNK2L5VV3WQEEIVNUHCQ5SV7B/graph.json","fetch_events":"https://pith.science/api/pith-number/YPNK2L5VV3WQEEIVNUHCQ5SV7B/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/YPNK2L5VV3WQEEIVNUHCQ5SV7B/action/timestamp_anchor","attest_storage":"https://pith.science/pith/YPNK2L5VV3WQEEIVNUHCQ5SV7B/action/storage_attestation","attest_author":"https://pith.science/pith/YPNK2L5VV3WQEEIVNUHCQ5SV7B/action/author_attestation","sign_citation":"https://pith.science/pith/YPNK2L5VV3WQEEIVNUHCQ5SV7B/action/citation_signature","submit_replication":"https://pith.science/pith/YPNK2L5VV3WQEEIVNUHCQ5SV7B/action/replication_record"}},"created_at":"2026-07-05T09:15:15.220811+00:00","updated_at":"2026-07-05T09:15:15.220811+00:00"}