{"state_type":"pith_open_graph_state","state_version":"1.0","pith_number":"pith:2024:CKQSJSYXKAYGDX6EIU3EIOSV2E","merge_version":"pith-open-graph-merge-v1","event_count":2,"valid_event_count":2,"invalid_event_count":0,"equivocation_count":0,"current":{"canonical_record":{"metadata":{"abstract_canon_sha256":"2483e31e651042a2060498bdf47b4980cad60624fef2c3c2d56e03d2cb15ce13","cross_cats_sorted":["cs.CL"],"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2024-01-11T18:57:12Z","title_canon_sha256":"edb1708c592326bd601188ead943b4eab018c8639fed0b61984ea5e117188084"},"schema_version":"1.0","source":{"id":"2401.06121","kind":"arxiv","version":1}},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2401.06121","created_at":"2026-05-17T23:38:48Z"},{"alias_kind":"arxiv_version","alias_value":"2401.06121v1","created_at":"2026-05-17T23:38:48Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2401.06121","created_at":"2026-05-17T23:38:48Z"},{"alias_kind":"pith_short_12","alias_value":"CKQSJSYXKAYG","created_at":"2026-05-18T12:33:37Z"},{"alias_kind":"pith_short_16","alias_value":"CKQSJSYXKAYGDX6E","created_at":"2026-05-18T12:33:37Z"},{"alias_kind":"pith_short_8","alias_value":"CKQSJSYX","created_at":"2026-05-18T12:33:37Z"}],"graph_snapshots":[{"event_id":"sha256:99ceb13979fc7a4f1c4ead38be3a88c5bf1499361069d134c2e177cc0f9f7a64","target":"graph","created_at":"2026-05-17T23:38:48Z","signer":{"key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signer_id":"pith.science","signer_type":"pith_registry"},"payload":{"graph_snapshot":{"author_claims":{"count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57","strong_count":0},"builder_version":"pith-number-builder-2026-05-17-v1","claims":{"count":4,"items":[{"attestation":"unclaimed","claim_id":"C1","kind":"strongest_claim","source":"verdict.strongest_claim","status":"machine_extracted","text":"Importantly, none of the baselines we consider show effective unlearning motivating continued efforts to develop approaches for unlearning that effectively tune models so that they truly behave as if they were never trained on the forget data at all."},{"attestation":"unclaimed","claim_id":"C2","kind":"weakest_assumption","source":"verdict.weakest_assumption","status":"machine_extracted","text":"The assumption that results on synthetic fictitious author profiles will generalize to the difficulty of unlearning real sensitive information from actual large-scale training corpora."},{"attestation":"unclaimed","claim_id":"C3","kind":"one_line_summary","source":"verdict.one_line_summary","status":"machine_extracted","text":"TOFU is a new benchmark with synthetic profiles and metrics demonstrating that existing unlearning algorithms for LLMs fail to achieve effective forgetting of targeted information."},{"attestation":"unclaimed","claim_id":"C4","kind":"headline","source":"verdict.pith_extraction.headline","status":"machine_extracted","text":"Unlearning methods for large language models fail to make them behave as if specific training data was never seen."}],"snapshot_sha256":"50c9f0eeff59d3ad6db1367425e43caeabfb4a4a28c4e8b1708a0fd3f81d14f5"},"formal_canon":{"evidence_count":2,"snapshot_sha256":"d0cee415386baf0819dbc07dd821e319a7c65dab5dc2d54a9a14bf217d426316"},"paper":{"abstract_excerpt":"Large language models trained on massive corpora of data from the web can memorize and reproduce sensitive or private data raising both legal and ethical concerns. Unlearning, or tuning models to forget information present in their training data, provides us with a way to protect private data after training. Although several methods exist for such unlearning, it is unclear to what extent they result in models equivalent to those where the data to be forgotten was never learned in the first place. To address this challenge, we present TOFU, a Task of Fictitious Unlearning, as a benchmark aimed ","authors_text":"Avi Schwarzschild, J. Zico Kolter, Pratyush Maini, Zachary C. Lipton, Zhili Feng","cross_cats":["cs.CL"],"headline":"Unlearning methods for large language models fail to make them behave as if specific training data was never seen.","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2024-01-11T18:57:12Z","title":"TOFU: A Task of Fictitious Unlearning for LLMs"},"references":{"count":44,"internal_anchors":6,"resolved_work":44,"sample":[{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":1,"title":"Machine unlearning","work_id":"4d425956-f000-49d7-9d36-97aa726d7a61","year":2021},{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":2,"title":"Extracting training data from large language models","work_id":"ed991696-818d-409a-87e0-a4c7da18320b","year":2021},{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":3,"title":"Membership inference attacks from first principles","work_id":"0ffe7056-ca0e-4fcf-81a6-073d0c49bd83","year":2022},{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":4,"title":"Unlearn what you want to forget: Efficient unlearning for llms, 2023","work_id":"74bd5caf-472e-4e1e-8c53-267458d855fa","year":2023},{"cited_arxiv_id":"","doi":"10.3115/v1/w14-4012","is_internal_anchor":false,"ref_index":5,"title":"On the properties of neural machine translation: Encoder-decoder approaches","work_id":"62cd99e9-1990-4d03-bfe7-00e9b69af44b","year":2014}],"snapshot_sha256":"86b2914ab211ef364f30265387a40383fb0c056c921ac98548406c2fe46ed22d"},"source":{"id":"2401.06121","kind":"arxiv","version":1},"verdict":{"created_at":"2026-05-16T11:03:31.010641Z","id":"f24f607f-f42e-49bb-919f-ba54b0e6714d","model_set":{"reader":"grok-4.3"},"one_line_summary":"TOFU is a new benchmark with synthetic profiles and metrics demonstrating that existing unlearning algorithms for LLMs fail to achieve effective forgetting of targeted information.","pipeline_version":"pith-pipeline@v0.9.0","pith_extraction_headline":"Unlearning methods for large language models fail to make them behave as if specific training data was never seen.","strongest_claim":"Importantly, none of the baselines we consider show effective unlearning motivating continued efforts to develop approaches for unlearning that effectively tune models so that they truly behave as if they were never trained on the forget data at all.","weakest_assumption":"The assumption that results on synthetic fictitious author profiles will generalize to the difficulty of unlearning real sensitive information from actual large-scale training corpora."}},"verdict_id":"f24f607f-f42e-49bb-919f-ba54b0e6714d"}}],"author_attestations":[],"timestamp_anchors":[],"storage_attestations":[],"citation_signatures":[],"replication_records":[],"corrections":[],"mirror_hints":[],"record_created":{"event_id":"sha256:d0b964dabd909f81760daf36509533c18b9f5fcd68ab6b071528a7022eca6bbe","target":"record","created_at":"2026-05-17T23:38:48Z","signer":{"key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signer_id":"pith.science","signer_type":"pith_registry"},"payload":{"attestation_state":"computed","canonical_record":{"metadata":{"abstract_canon_sha256":"2483e31e651042a2060498bdf47b4980cad60624fef2c3c2d56e03d2cb15ce13","cross_cats_sorted":["cs.CL"],"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2024-01-11T18:57:12Z","title_canon_sha256":"edb1708c592326bd601188ead943b4eab018c8639fed0b61984ea5e117188084"},"schema_version":"1.0","source":{"id":"2401.06121","kind":"arxiv","version":1}},"canonical_sha256":"12a124cb17503061dfc44536443a55d12e762122b897a676bc828a2646543932","receipt":{"algorithm":"ed25519","builder_version":"pith-number-builder-2026-05-17-v1","canonical_sha256":"12a124cb17503061dfc44536443a55d12e762122b897a676bc828a2646543932","first_computed_at":"2026-05-17T23:38:48.074436Z","key_id":"pith-v1-2026-05","kind":"pith_receipt","last_reissued_at":"2026-05-17T23:38:48.074436Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","receipt_version":"0.3","signature_b64":"MKA/crtRZwidg5Vfirl0xQcuyJxNqTw5g2APt8KwSKgxrpaMRsW8da4kWvi75eLt3qQNSqtYzdgCYSLAyGZiCw==","signature_status":"signed_v1","signed_at":"2026-05-17T23:38:48.074970Z","signed_message":"canonical_sha256_bytes"},"source_id":"2401.06121","source_kind":"arxiv","source_version":1}}},"equivocations":[],"invalid_events":[],"applied_event_ids":["sha256:d0b964dabd909f81760daf36509533c18b9f5fcd68ab6b071528a7022eca6bbe","sha256:99ceb13979fc7a4f1c4ead38be3a88c5bf1499361069d134c2e177cc0f9f7a64"],"state_sha256":"58581552e06d728e1fa2301837a03a03d547982d6b1ec0f06bd498383bfdd124"}