{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2022:WWLMW464XEPKA56ATZBMFD4PXR","short_pith_number":"pith:WWLMW464","schema_version":"1.0","canonical_sha256":"b596cb73dcb91ea077c09e42c28f8fbc6461370149597fcc1e47d6c6d784672f","source":{"kind":"arxiv","id":"2205.13636","version":2},"attestation_state":"computed","paper":{"title":"Quark: Controllable Text Generation with Reinforced Unlearning","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.LG"],"primary_cat":"cs.CL","authors_text":"Jack Hessel, Lianhui Qin, Liwei Jiang, Peter West, Prithviraj Ammanabrolu, Sean Welleck, Ximing Lu, Yejin Choi","submitted_at":"2022-05-26T21:11:51Z","abstract_excerpt":"Large-scale language models often learn behaviors that are misaligned with user expectations. Generated text may contain offensive or toxic language, contain significant repetition, or be of a different sentiment than desired by the user. We consider the task of unlearning these misalignments by fine-tuning the language model on signals of what not to do. We introduce Quantized Reward Konditioning (Quark), an algorithm for optimizing a reward function that quantifies an (un)wanted property, while not straying too far from the original model. Quark alternates between (i) collecting samples with"},"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":"2205.13636","kind":"arxiv","version":2},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CL","submitted_at":"2022-05-26T21:11:51Z","cross_cats_sorted":["cs.LG"],"title_canon_sha256":"8edf80a907a608ca612b39e26e62d652c67e97dabec4bc99bc1701f2e214c143","abstract_canon_sha256":"4870bd0c82882fdf9750f60089ed631ea71477bd0bad8171d92b54866e849086"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T05:16:51.047226Z","signature_b64":"m5oIUWS+r+Bj6ksHBb40ifZrR2gxRhnff5hDH3+OloKhvATVxDom6HCvVarQZdgLKiZX8N+TFoGLsPinJmf/AQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"b596cb73dcb91ea077c09e42c28f8fbc6461370149597fcc1e47d6c6d784672f","last_reissued_at":"2026-07-05T05:16:51.046781Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T05:16:51.046781Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Quark: Controllable Text Generation with Reinforced Unlearning","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.LG"],"primary_cat":"cs.CL","authors_text":"Jack Hessel, Lianhui Qin, Liwei Jiang, Peter West, Prithviraj Ammanabrolu, Sean Welleck, Ximing Lu, Yejin Choi","submitted_at":"2022-05-26T21:11:51Z","abstract_excerpt":"Large-scale language models often learn behaviors that are misaligned with user expectations. Generated text may contain offensive or toxic language, contain significant repetition, or be of a different sentiment than desired by the user. We consider the task of unlearning these misalignments by fine-tuning the language model on signals of what not to do. We introduce Quantized Reward Konditioning (Quark), an algorithm for optimizing a reward function that quantifies an (un)wanted property, while not straying too far from the original model. Quark alternates between (i) collecting samples with"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2205.13636","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/2205.13636/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":"2205.13636","created_at":"2026-07-05T05:16:51.046842+00:00"},{"alias_kind":"arxiv_version","alias_value":"2205.13636v2","created_at":"2026-07-05T05:16:51.046842+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2205.13636","created_at":"2026-07-05T05:16:51.046842+00:00"},{"alias_kind":"pith_short_12","alias_value":"WWLMW464XEPK","created_at":"2026-07-05T05:16:51.046842+00:00"},{"alias_kind":"pith_short_16","alias_value":"WWLMW464XEPKA56A","created_at":"2026-07-05T05:16:51.046842+00:00"},{"alias_kind":"pith_short_8","alias_value":"WWLMW464","created_at":"2026-07-05T05:16:51.046842+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":4,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2402.05070","citing_title":"A Roadmap to Pluralistic Alignment","ref_index":297,"is_internal_anchor":false},{"citing_arxiv_id":"2305.11738","citing_title":"CRITIC: Large Language Models Can Self-Correct with Tool-Interactive Critiquing","ref_index":2,"is_internal_anchor":false},{"citing_arxiv_id":"2212.04089","citing_title":"Editing Models with Task Arithmetic","ref_index":64,"is_internal_anchor":false},{"citing_arxiv_id":"2303.17651","citing_title":"Self-Refine: Iterative Refinement with Self-Feedback","ref_index":22,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/WWLMW464XEPKA56ATZBMFD4PXR","json":"https://pith.science/pith/WWLMW464XEPKA56ATZBMFD4PXR.json","graph_json":"https://pith.science/api/pith-number/WWLMW464XEPKA56ATZBMFD4PXR/graph.json","events_json":"https://pith.science/api/pith-number/WWLMW464XEPKA56ATZBMFD4PXR/events.json","paper":"https://pith.science/paper/WWLMW464"},"agent_actions":{"view_html":"https://pith.science/pith/WWLMW464XEPKA56ATZBMFD4PXR","download_json":"https://pith.science/pith/WWLMW464XEPKA56ATZBMFD4PXR.json","view_paper":"https://pith.science/paper/WWLMW464","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2205.13636&json=true","fetch_graph":"https://pith.science/api/pith-number/WWLMW464XEPKA56ATZBMFD4PXR/graph.json","fetch_events":"https://pith.science/api/pith-number/WWLMW464XEPKA56ATZBMFD4PXR/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/WWLMW464XEPKA56ATZBMFD4PXR/action/timestamp_anchor","attest_storage":"https://pith.science/pith/WWLMW464XEPKA56ATZBMFD4PXR/action/storage_attestation","attest_author":"https://pith.science/pith/WWLMW464XEPKA56ATZBMFD4PXR/action/author_attestation","sign_citation":"https://pith.science/pith/WWLMW464XEPKA56ATZBMFD4PXR/action/citation_signature","submit_replication":"https://pith.science/pith/WWLMW464XEPKA56ATZBMFD4PXR/action/replication_record"}},"created_at":"2026-07-05T05:16:51.046842+00:00","updated_at":"2026-07-05T05:16:51.046842+00:00"}