{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2024:R6CRE3WVTBRYJEJGX3EBLPAR3P","short_pith_number":"pith:R6CRE3WV","schema_version":"1.0","canonical_sha256":"8f85126ed59863849126bec815bc11dbe77c5c8950114caa49d89daeb4ed20b9","source":{"kind":"arxiv","id":"2404.10464","version":3},"attestation_state":"computed","paper":{"title":"DESTEIN: Navigating Detoxification of Language Models via Universal Steering Pairs and Head-wise Activation Fusion","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.CL","authors_text":"Chuanyang Gong, Han Jiang, Yu Li, Zhihua Wei","submitted_at":"2024-04-16T11:07:48Z","abstract_excerpt":"Despite the remarkable achievements of language models (LMs) across a broad spectrum of tasks, their propensity for generating toxic outputs remains a prevalent concern. Current solutions involving finetuning or auxiliary models usually require extensive computational resources, hindering their practicality in large language models (LLMs). In this paper, we propose DeStein, a novel method that detoxifies LMs by applying representation engineering in activation spaces with lower resource and time costs. Specifically, we derive detoxification vectors from self-induced, universal steering pairs t"},"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":"2404.10464","kind":"arxiv","version":3},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CL","submitted_at":"2024-04-16T11:07:48Z","cross_cats_sorted":["cs.AI"],"title_canon_sha256":"dd128dcc29e70fc2c4c53544385e43461e683b5ed48205ce71fc9a3cac926c01","abstract_canon_sha256":"f7411a882f74cac85f287c7a0524116308e8e1d87a2a16f467dcff62b439ed91"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T08:54:10.224521Z","signature_b64":"asax42mwoaTe8Y+fpBHWGgKYMXUfa9VRcNIMos01cvyoqSYrMB15I1xQk0B2wFXF/iU5k2vAA8qZ0MP67+6JCQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"8f85126ed59863849126bec815bc11dbe77c5c8950114caa49d89daeb4ed20b9","last_reissued_at":"2026-07-05T08:54:10.224081Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T08:54:10.224081Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"DESTEIN: Navigating Detoxification of Language Models via Universal Steering Pairs and Head-wise Activation Fusion","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.CL","authors_text":"Chuanyang Gong, Han Jiang, Yu Li, Zhihua Wei","submitted_at":"2024-04-16T11:07:48Z","abstract_excerpt":"Despite the remarkable achievements of language models (LMs) across a broad spectrum of tasks, their propensity for generating toxic outputs remains a prevalent concern. Current solutions involving finetuning or auxiliary models usually require extensive computational resources, hindering their practicality in large language models (LLMs). In this paper, we propose DeStein, a novel method that detoxifies LMs by applying representation engineering in activation spaces with lower resource and time costs. Specifically, we derive detoxification vectors from self-induced, universal steering pairs t"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2404.10464","kind":"arxiv","version":3},"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/2404.10464/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":"2404.10464","created_at":"2026-07-05T08:54:10.224143+00:00"},{"alias_kind":"arxiv_version","alias_value":"2404.10464v3","created_at":"2026-07-05T08:54:10.224143+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2404.10464","created_at":"2026-07-05T08:54:10.224143+00:00"},{"alias_kind":"pith_short_12","alias_value":"R6CRE3WVTBRY","created_at":"2026-07-05T08:54:10.224143+00:00"},{"alias_kind":"pith_short_16","alias_value":"R6CRE3WVTBRYJEJG","created_at":"2026-07-05T08:54:10.224143+00:00"},{"alias_kind":"pith_short_8","alias_value":"R6CRE3WV","created_at":"2026-07-05T08:54:10.224143+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":1,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2507.03724","citing_title":"MemOS: A Memory OS for AI System","ref_index":44,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/R6CRE3WVTBRYJEJGX3EBLPAR3P","json":"https://pith.science/pith/R6CRE3WVTBRYJEJGX3EBLPAR3P.json","graph_json":"https://pith.science/api/pith-number/R6CRE3WVTBRYJEJGX3EBLPAR3P/graph.json","events_json":"https://pith.science/api/pith-number/R6CRE3WVTBRYJEJGX3EBLPAR3P/events.json","paper":"https://pith.science/paper/R6CRE3WV"},"agent_actions":{"view_html":"https://pith.science/pith/R6CRE3WVTBRYJEJGX3EBLPAR3P","download_json":"https://pith.science/pith/R6CRE3WVTBRYJEJGX3EBLPAR3P.json","view_paper":"https://pith.science/paper/R6CRE3WV","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2404.10464&json=true","fetch_graph":"https://pith.science/api/pith-number/R6CRE3WVTBRYJEJGX3EBLPAR3P/graph.json","fetch_events":"https://pith.science/api/pith-number/R6CRE3WVTBRYJEJGX3EBLPAR3P/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/R6CRE3WVTBRYJEJGX3EBLPAR3P/action/timestamp_anchor","attest_storage":"https://pith.science/pith/R6CRE3WVTBRYJEJGX3EBLPAR3P/action/storage_attestation","attest_author":"https://pith.science/pith/R6CRE3WVTBRYJEJGX3EBLPAR3P/action/author_attestation","sign_citation":"https://pith.science/pith/R6CRE3WVTBRYJEJGX3EBLPAR3P/action/citation_signature","submit_replication":"https://pith.science/pith/R6CRE3WVTBRYJEJGX3EBLPAR3P/action/replication_record"}},"created_at":"2026-07-05T08:54:10.224143+00:00","updated_at":"2026-07-05T08:54:10.224143+00:00"}