{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2022:ISPLZLNKWRQBDZI5ZB6YAMBHZU","short_pith_number":"pith:ISPLZLNK","schema_version":"1.0","canonical_sha256":"449ebcadaab46011e51dc87d803027cd3c08a975498a379e8da78a76c6ada37b","source":{"kind":"arxiv","id":"2203.03072","version":1},"attestation_state":"computed","paper":{"title":"Leashing the Inner Demons: Self-Detoxification for Language Models","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.AI","cs.LG"],"primary_cat":"cs.CL","authors_text":"Canwen Xu, Julian McAuley, Zexue He, Zhankui He","submitted_at":"2022-03-06T23:55:12Z","abstract_excerpt":"Language models (LMs) can reproduce (or amplify) toxic language seen during training, which poses a risk to their practical application. In this paper, we conduct extensive experiments to study this phenomenon. We analyze the impact of prompts, decoding strategies and training corpora on the output toxicity. Based on our findings, we propose a simple yet effective method for language models to \"detoxify\" themselves without an additional large corpus or external discriminator. Compared to a supervised baseline, our proposed method shows better toxicity reduction with good generation quality in "},"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":"2203.03072","kind":"arxiv","version":1},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CL","submitted_at":"2022-03-06T23:55:12Z","cross_cats_sorted":["cs.AI","cs.LG"],"title_canon_sha256":"a4352a0e2b7eda7fad4d2680487b895bb49536283dbe47e00d17dd370f911f9a","abstract_canon_sha256":"8a119df5f689e6d6a3c59acab4ab4ec7d82b66d84016c8e75bc4f4ece06a90b8"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T04:02:30.933188Z","signature_b64":"myv+eRgMGoSS4dzDO7ElPaVScIACUzrxrA2w7rdb1Jn300xtB5svJ3q7pOmrrpKG7qCXS8doSXpnA6rLvLiCCQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"449ebcadaab46011e51dc87d803027cd3c08a975498a379e8da78a76c6ada37b","last_reissued_at":"2026-07-05T04:02:30.932795Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T04:02:30.932795Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Leashing the Inner Demons: Self-Detoxification for Language Models","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.AI","cs.LG"],"primary_cat":"cs.CL","authors_text":"Canwen Xu, Julian McAuley, Zexue He, Zhankui He","submitted_at":"2022-03-06T23:55:12Z","abstract_excerpt":"Language models (LMs) can reproduce (or amplify) toxic language seen during training, which poses a risk to their practical application. In this paper, we conduct extensive experiments to study this phenomenon. We analyze the impact of prompts, decoding strategies and training corpora on the output toxicity. Based on our findings, we propose a simple yet effective method for language models to \"detoxify\" themselves without an additional large corpus or external discriminator. Compared to a supervised baseline, our proposed method shows better toxicity reduction with good generation quality in "},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2203.03072","kind":"arxiv","version":1},"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/2203.03072/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":"2203.03072","created_at":"2026-07-05T04:02:30.932851+00:00"},{"alias_kind":"arxiv_version","alias_value":"2203.03072v1","created_at":"2026-07-05T04:02:30.932851+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2203.03072","created_at":"2026-07-05T04:02:30.932851+00:00"},{"alias_kind":"pith_short_12","alias_value":"ISPLZLNKWRQB","created_at":"2026-07-05T04:02:30.932851+00:00"},{"alias_kind":"pith_short_16","alias_value":"ISPLZLNKWRQBDZI5","created_at":"2026-07-05T04:02:30.932851+00:00"},{"alias_kind":"pith_short_8","alias_value":"ISPLZLNK","created_at":"2026-07-05T04:02:30.932851+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/ISPLZLNKWRQBDZI5ZB6YAMBHZU","json":"https://pith.science/pith/ISPLZLNKWRQBDZI5ZB6YAMBHZU.json","graph_json":"https://pith.science/api/pith-number/ISPLZLNKWRQBDZI5ZB6YAMBHZU/graph.json","events_json":"https://pith.science/api/pith-number/ISPLZLNKWRQBDZI5ZB6YAMBHZU/events.json","paper":"https://pith.science/paper/ISPLZLNK"},"agent_actions":{"view_html":"https://pith.science/pith/ISPLZLNKWRQBDZI5ZB6YAMBHZU","download_json":"https://pith.science/pith/ISPLZLNKWRQBDZI5ZB6YAMBHZU.json","view_paper":"https://pith.science/paper/ISPLZLNK","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2203.03072&json=true","fetch_graph":"https://pith.science/api/pith-number/ISPLZLNKWRQBDZI5ZB6YAMBHZU/graph.json","fetch_events":"https://pith.science/api/pith-number/ISPLZLNKWRQBDZI5ZB6YAMBHZU/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/ISPLZLNKWRQBDZI5ZB6YAMBHZU/action/timestamp_anchor","attest_storage":"https://pith.science/pith/ISPLZLNKWRQBDZI5ZB6YAMBHZU/action/storage_attestation","attest_author":"https://pith.science/pith/ISPLZLNKWRQBDZI5ZB6YAMBHZU/action/author_attestation","sign_citation":"https://pith.science/pith/ISPLZLNKWRQBDZI5ZB6YAMBHZU/action/citation_signature","submit_replication":"https://pith.science/pith/ISPLZLNKWRQBDZI5ZB6YAMBHZU/action/replication_record"}},"created_at":"2026-07-05T04:02:30.932851+00:00","updated_at":"2026-07-05T04:02:30.932851+00:00"}