{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2023:DBWPQYME3IAYGZA3J32DMJF6BG","short_pith_number":"pith:DBWPQYME","schema_version":"1.0","canonical_sha256":"186cf86184da0183641b4ef43624be0980a89b60c5121956ac12e18e5c50a28c","source":{"kind":"arxiv","id":"2309.07124","version":2},"attestation_state":"computed","paper":{"title":"RAIN: Your Language Models Can Align Themselves without Finetuning","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Chao Zhang, Fangyun Wei, Hongyang Zhang, Jinjing Zhao, Yuhui Li","submitted_at":"2023-09-13T17:59:09Z","abstract_excerpt":"Large language models (LLMs) often demonstrate inconsistencies with human preferences. Previous research typically gathered human preference data and then aligned the pre-trained models using reinforcement learning or instruction tuning, a.k.a. the finetuning step. In contrast, aligning frozen LLMs without requiring alignment data is more appealing. This work explores the potential of the latter setting. We discover that by integrating self-evaluation and rewind mechanisms, unaligned LLMs can directly produce responses consistent with human preferences via self-boosting. We introduce a novel i"},"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":"2309.07124","kind":"arxiv","version":2},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CL","submitted_at":"2023-09-13T17:59:09Z","cross_cats_sorted":[],"title_canon_sha256":"dc8f2fd814a53ca3ed0a243b6d752d8c6ae67a1db6c0c4fbb7f91800f7d2d44b","abstract_canon_sha256":"1fb0ada953e9b8a0de6822ebf538c4ff22b9588642b6525fcde9860dfd881dc5"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T06:58:42.481122Z","signature_b64":"Ownh56ZBZBAuRj2hLPnCLdgdHUSUqmcaeTxGleczlfLEEcL5/NmjA0WtywOpoTOZk70PCJ7ZD0kmNtflVdizBQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"186cf86184da0183641b4ef43624be0980a89b60c5121956ac12e18e5c50a28c","last_reissued_at":"2026-07-05T06:58:42.480606Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T06:58:42.480606Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"RAIN: Your Language Models Can Align Themselves without Finetuning","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Chao Zhang, Fangyun Wei, Hongyang Zhang, Jinjing Zhao, Yuhui Li","submitted_at":"2023-09-13T17:59:09Z","abstract_excerpt":"Large language models (LLMs) often demonstrate inconsistencies with human preferences. Previous research typically gathered human preference data and then aligned the pre-trained models using reinforcement learning or instruction tuning, a.k.a. the finetuning step. In contrast, aligning frozen LLMs without requiring alignment data is more appealing. This work explores the potential of the latter setting. We discover that by integrating self-evaluation and rewind mechanisms, unaligned LLMs can directly produce responses consistent with human preferences via self-boosting. We introduce a novel i"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2309.07124","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/2309.07124/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":"2309.07124","created_at":"2026-07-05T06:58:42.480674+00:00"},{"alias_kind":"arxiv_version","alias_value":"2309.07124v2","created_at":"2026-07-05T06:58:42.480674+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2309.07124","created_at":"2026-07-05T06:58:42.480674+00:00"},{"alias_kind":"pith_short_12","alias_value":"DBWPQYME3IAY","created_at":"2026-07-05T06:58:42.480674+00:00"},{"alias_kind":"pith_short_16","alias_value":"DBWPQYME3IAYGZA3","created_at":"2026-07-05T06:58:42.480674+00:00"},{"alias_kind":"pith_short_8","alias_value":"DBWPQYME","created_at":"2026-07-05T06:58:42.480674+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":5,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.10651","citing_title":"Kwai Keye-VL-2.0 Technical Report","ref_index":60,"is_internal_anchor":false},{"citing_arxiv_id":"2505.19075","citing_title":"Universal Reasoner: A Single, Composable Plug-and-Play Reasoner for Frozen LLMs","ref_index":21,"is_internal_anchor":false},{"citing_arxiv_id":"2506.00166","citing_title":"Disentangled Safety Adapters Enable Efficient Guardrails and Flexible Inference-Time Alignment","ref_index":25,"is_internal_anchor":false},{"citing_arxiv_id":"2310.02446","citing_title":"Low-Resource Languages Jailbreak GPT-4","ref_index":27,"is_internal_anchor":false},{"citing_arxiv_id":"2407.04295","citing_title":"Jailbreak Attacks and Defenses Against Large Language Models: A Survey","ref_index":53,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/DBWPQYME3IAYGZA3J32DMJF6BG","json":"https://pith.science/pith/DBWPQYME3IAYGZA3J32DMJF6BG.json","graph_json":"https://pith.science/api/pith-number/DBWPQYME3IAYGZA3J32DMJF6BG/graph.json","events_json":"https://pith.science/api/pith-number/DBWPQYME3IAYGZA3J32DMJF6BG/events.json","paper":"https://pith.science/paper/DBWPQYME"},"agent_actions":{"view_html":"https://pith.science/pith/DBWPQYME3IAYGZA3J32DMJF6BG","download_json":"https://pith.science/pith/DBWPQYME3IAYGZA3J32DMJF6BG.json","view_paper":"https://pith.science/paper/DBWPQYME","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2309.07124&json=true","fetch_graph":"https://pith.science/api/pith-number/DBWPQYME3IAYGZA3J32DMJF6BG/graph.json","fetch_events":"https://pith.science/api/pith-number/DBWPQYME3IAYGZA3J32DMJF6BG/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/DBWPQYME3IAYGZA3J32DMJF6BG/action/timestamp_anchor","attest_storage":"https://pith.science/pith/DBWPQYME3IAYGZA3J32DMJF6BG/action/storage_attestation","attest_author":"https://pith.science/pith/DBWPQYME3IAYGZA3J32DMJF6BG/action/author_attestation","sign_citation":"https://pith.science/pith/DBWPQYME3IAYGZA3J32DMJF6BG/action/citation_signature","submit_replication":"https://pith.science/pith/DBWPQYME3IAYGZA3J32DMJF6BG/action/replication_record"}},"created_at":"2026-07-05T06:58:42.480674+00:00","updated_at":"2026-07-05T06:58:42.480674+00:00"}