{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2024:NUKS2PFWSNY54SPWFDJDVHY2UD","short_pith_number":"pith:NUKS2PFW","schema_version":"1.0","canonical_sha256":"6d152d3cb69371de49f628d23a9f1aa0f4df63c3f37372b19fad1b6266de97e6","source":{"kind":"arxiv","id":"2406.02721","version":3},"attestation_state":"computed","paper":{"title":"Self-Control of LLM Behaviors by Compressing Suffix Gradient into Prefix Controller","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.CL","authors_text":"Dan Zhang, Difan Zou, Fan Yin, Min Cai, Shichang Zhang, Yisong Yue, Yuchen Zhang, Ziniu Hu","submitted_at":"2024-06-04T19:05:10Z","abstract_excerpt":"We propose SelfControl, an inference-time model control method utilizing gradients to control the behavior of large language models (LLMs) without explicit human annotations. Given a desired behavior expressed in a natural language suffix string concatenated to the input prompt, SelfControl computes gradients of the LLM's self-evaluation of the suffix with respect to its latent representations. The gradients are used to directly control the auto-regressive generation process towards desired behaviors, which eliminates human supervision, achieves precise and transparent control, and offers on-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":"2406.02721","kind":"arxiv","version":3},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CL","submitted_at":"2024-06-04T19:05:10Z","cross_cats_sorted":["cs.AI"],"title_canon_sha256":"1aece65f59d09ed87e4786440f871986dac281b8975bfe73eca8afd18d4b216e","abstract_canon_sha256":"cfcc597e564e60daf1746be001d1ad5e5633002b3e5caeef310b24aa53f93517"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T09:19:28.850850Z","signature_b64":"nIXxCPU7x9FW3r+flXtz+Q+Kp9AVHTmwEnXvCuAe2IHZKCZntLMCZUz/EU72s+X2Ns+Vpcdb/634h47O6KXvAQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"6d152d3cb69371de49f628d23a9f1aa0f4df63c3f37372b19fad1b6266de97e6","last_reissued_at":"2026-07-05T09:19:28.850448Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T09:19:28.850448Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Self-Control of LLM Behaviors by Compressing Suffix Gradient into Prefix Controller","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.CL","authors_text":"Dan Zhang, Difan Zou, Fan Yin, Min Cai, Shichang Zhang, Yisong Yue, Yuchen Zhang, Ziniu Hu","submitted_at":"2024-06-04T19:05:10Z","abstract_excerpt":"We propose SelfControl, an inference-time model control method utilizing gradients to control the behavior of large language models (LLMs) without explicit human annotations. Given a desired behavior expressed in a natural language suffix string concatenated to the input prompt, SelfControl computes gradients of the LLM's self-evaluation of the suffix with respect to its latent representations. The gradients are used to directly control the auto-regressive generation process towards desired behaviors, which eliminates human supervision, achieves precise and transparent control, and offers on-t"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2406.02721","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/2406.02721/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":"2406.02721","created_at":"2026-07-05T09:19:28.850497+00:00"},{"alias_kind":"arxiv_version","alias_value":"2406.02721v3","created_at":"2026-07-05T09:19:28.850497+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2406.02721","created_at":"2026-07-05T09:19:28.850497+00:00"},{"alias_kind":"pith_short_12","alias_value":"NUKS2PFWSNY5","created_at":"2026-07-05T09:19:28.850497+00:00"},{"alias_kind":"pith_short_16","alias_value":"NUKS2PFWSNY54SPW","created_at":"2026-07-05T09:19:28.850497+00:00"},{"alias_kind":"pith_short_8","alias_value":"NUKS2PFW","created_at":"2026-07-05T09:19:28.850497+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":1,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2512.05929","citing_title":"LLM Harms: A Taxonomy and Discussion","ref_index":213,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/NUKS2PFWSNY54SPWFDJDVHY2UD","json":"https://pith.science/pith/NUKS2PFWSNY54SPWFDJDVHY2UD.json","graph_json":"https://pith.science/api/pith-number/NUKS2PFWSNY54SPWFDJDVHY2UD/graph.json","events_json":"https://pith.science/api/pith-number/NUKS2PFWSNY54SPWFDJDVHY2UD/events.json","paper":"https://pith.science/paper/NUKS2PFW"},"agent_actions":{"view_html":"https://pith.science/pith/NUKS2PFWSNY54SPWFDJDVHY2UD","download_json":"https://pith.science/pith/NUKS2PFWSNY54SPWFDJDVHY2UD.json","view_paper":"https://pith.science/paper/NUKS2PFW","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2406.02721&json=true","fetch_graph":"https://pith.science/api/pith-number/NUKS2PFWSNY54SPWFDJDVHY2UD/graph.json","fetch_events":"https://pith.science/api/pith-number/NUKS2PFWSNY54SPWFDJDVHY2UD/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/NUKS2PFWSNY54SPWFDJDVHY2UD/action/timestamp_anchor","attest_storage":"https://pith.science/pith/NUKS2PFWSNY54SPWFDJDVHY2UD/action/storage_attestation","attest_author":"https://pith.science/pith/NUKS2PFWSNY54SPWFDJDVHY2UD/action/author_attestation","sign_citation":"https://pith.science/pith/NUKS2PFWSNY54SPWFDJDVHY2UD/action/citation_signature","submit_replication":"https://pith.science/pith/NUKS2PFWSNY54SPWFDJDVHY2UD/action/replication_record"}},"created_at":"2026-07-05T09:19:28.850497+00:00","updated_at":"2026-07-05T09:19:28.850497+00:00"}