{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2024:V56BPYPBWM36VOZAZWOPW6CWRI","short_pith_number":"pith:V56BPYPB","schema_version":"1.0","canonical_sha256":"af7c17e1e1b337eabb20cd9cfb78568a0d54e58cd10662a76c42ead0b0a769e7","source":{"kind":"arxiv","id":"2410.02879","version":2},"attestation_state":"computed","paper":{"title":"Position: LLM Unlearning Benchmarks are Weak Measures of Progress","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Neil Kale, Pratiksha Thaker, Shengyuan Hu, Virginia Smith, Yash Maurya, Zhiwei Steven Wu","submitted_at":"2024-10-03T18:07:25Z","abstract_excerpt":"Unlearning methods have the potential to improve the privacy and safety of large language models (LLMs) by removing sensitive or harmful information post hoc. The LLM unlearning research community has increasingly turned toward empirical benchmarks to assess the effectiveness of such methods. In this paper, we find that existing benchmarks provide an overly optimistic and potentially misleading view on the effectiveness of candidate unlearning methods. By introducing simple, benign modifications to a number of popular benchmarks, we expose instances where supposedly unlearned information remai"},"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":"2410.02879","kind":"arxiv","version":2},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CL","submitted_at":"2024-10-03T18:07:25Z","cross_cats_sorted":[],"title_canon_sha256":"ac167948a8fdfeb5406a7e51d20d309ab43a91be32ed9975436bc242b5b09f59","abstract_canon_sha256":"ef2e1be9711f2bc021f23a70789f0ec57b99a77a9e66693970e2f66497084936"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T10:45:55.887408Z","signature_b64":"12jZ53HqfsH/sDDrPjsugxZJBgv/6ctdrHW69HMi7bQZeSKGA923oOoa6UlRAwDTbPVZwpIdz438oL+Qii5fAg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"af7c17e1e1b337eabb20cd9cfb78568a0d54e58cd10662a76c42ead0b0a769e7","last_reissued_at":"2026-07-05T10:45:55.886825Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T10:45:55.886825Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Position: LLM Unlearning Benchmarks are Weak Measures of Progress","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Neil Kale, Pratiksha Thaker, Shengyuan Hu, Virginia Smith, Yash Maurya, Zhiwei Steven Wu","submitted_at":"2024-10-03T18:07:25Z","abstract_excerpt":"Unlearning methods have the potential to improve the privacy and safety of large language models (LLMs) by removing sensitive or harmful information post hoc. The LLM unlearning research community has increasingly turned toward empirical benchmarks to assess the effectiveness of such methods. In this paper, we find that existing benchmarks provide an overly optimistic and potentially misleading view on the effectiveness of candidate unlearning methods. By introducing simple, benign modifications to a number of popular benchmarks, we expose instances where supposedly unlearned information remai"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2410.02879","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/2410.02879/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":"2410.02879","created_at":"2026-07-05T10:45:55.886898+00:00"},{"alias_kind":"arxiv_version","alias_value":"2410.02879v2","created_at":"2026-07-05T10:45:55.886898+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2410.02879","created_at":"2026-07-05T10:45:55.886898+00:00"},{"alias_kind":"pith_short_12","alias_value":"V56BPYPBWM36","created_at":"2026-07-05T10:45:55.886898+00:00"},{"alias_kind":"pith_short_16","alias_value":"V56BPYPBWM36VOZA","created_at":"2026-07-05T10:45:55.886898+00:00"},{"alias_kind":"pith_short_8","alias_value":"V56BPYPB","created_at":"2026-07-05T10:45:55.886898+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":1,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.10989","citing_title":"Null-Space Constrained Low-Rank Adaptation for Response-Specified Large Language Model Unlearning","ref_index":8,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/V56BPYPBWM36VOZAZWOPW6CWRI","json":"https://pith.science/pith/V56BPYPBWM36VOZAZWOPW6CWRI.json","graph_json":"https://pith.science/api/pith-number/V56BPYPBWM36VOZAZWOPW6CWRI/graph.json","events_json":"https://pith.science/api/pith-number/V56BPYPBWM36VOZAZWOPW6CWRI/events.json","paper":"https://pith.science/paper/V56BPYPB"},"agent_actions":{"view_html":"https://pith.science/pith/V56BPYPBWM36VOZAZWOPW6CWRI","download_json":"https://pith.science/pith/V56BPYPBWM36VOZAZWOPW6CWRI.json","view_paper":"https://pith.science/paper/V56BPYPB","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2410.02879&json=true","fetch_graph":"https://pith.science/api/pith-number/V56BPYPBWM36VOZAZWOPW6CWRI/graph.json","fetch_events":"https://pith.science/api/pith-number/V56BPYPBWM36VOZAZWOPW6CWRI/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/V56BPYPBWM36VOZAZWOPW6CWRI/action/timestamp_anchor","attest_storage":"https://pith.science/pith/V56BPYPBWM36VOZAZWOPW6CWRI/action/storage_attestation","attest_author":"https://pith.science/pith/V56BPYPBWM36VOZAZWOPW6CWRI/action/author_attestation","sign_citation":"https://pith.science/pith/V56BPYPBWM36VOZAZWOPW6CWRI/action/citation_signature","submit_replication":"https://pith.science/pith/V56BPYPBWM36VOZAZWOPW6CWRI/action/replication_record"}},"created_at":"2026-07-05T10:45:55.886898+00:00","updated_at":"2026-07-05T10:45:55.886898+00:00"}