{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:75OYLG6NPQSBPO2QLDEYQ5ATNU","short_pith_number":"pith:75OYLG6N","schema_version":"1.0","canonical_sha256":"ff5d859bcd7c2417bb5058c98874136d3aa461b3f555ca1ad35b7b7787d5bc4b","source":{"kind":"arxiv","id":"2510.02453","version":3},"attestation_state":"computed","paper":{"title":"How to Train Your Advisor: Steering Black-Box LLMs with Advisor Models","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI","cs.CL"],"primary_cat":"cs.LG","authors_text":"Abigail O'Neill, Alan Zhu, Alexandros G. Dimakis, Joseph E. Gonzalez, Matei Zaharia, Parth Asawa","submitted_at":"2025-10-02T18:02:39Z","abstract_excerpt":"Frontier language models are deployed as black-box services, where model weights cannot be modified and customization is limited to prompting. We introduce Advisor Models, a method to train small open-weight models to generate dynamic, per-instance natural language advice that improves the capabilities of black-box frontier models. Advisor Models improve GPT-5.2's performance on RuleArena (Taxes) by 27.4%, reduce Gemini 3 Pro's steps taken in SWE agent tasks by 24.6%, and outperform static prompt optimizers in personalizing GPT-5 to user preferences (85-100% vs. 40-60%). We also find that advi"},"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":"2510.02453","kind":"arxiv","version":3},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2025-10-02T18:02:39Z","cross_cats_sorted":["cs.AI","cs.CL"],"title_canon_sha256":"0d946386449a49411123223c0a31810b3c788b3c523059ed750e9baf89e88f7c","abstract_canon_sha256":"5396dfd1cb06015ad0d7b50e414ad102b53da60ef837bb603a365977bc71ca1d"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-05-20T00:01:35.833585Z","signature_b64":"jQfSQnjmoCFjvhbP1knwp9AbNoUlhTIMHwQDtQYnElfAv7CjrgqoEafpkyVJ+mN5wUbyDv0twyOadmzaPNeFAQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"ff5d859bcd7c2417bb5058c98874136d3aa461b3f555ca1ad35b7b7787d5bc4b","last_reissued_at":"2026-05-20T00:01:35.832777Z","signature_status":"signed_v1","first_computed_at":"2026-05-20T00:01:35.832777Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"How to Train Your Advisor: Steering Black-Box LLMs with Advisor Models","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI","cs.CL"],"primary_cat":"cs.LG","authors_text":"Abigail O'Neill, Alan Zhu, Alexandros G. Dimakis, Joseph E. Gonzalez, Matei Zaharia, Parth Asawa","submitted_at":"2025-10-02T18:02:39Z","abstract_excerpt":"Frontier language models are deployed as black-box services, where model weights cannot be modified and customization is limited to prompting. We introduce Advisor Models, a method to train small open-weight models to generate dynamic, per-instance natural language advice that improves the capabilities of black-box frontier models. Advisor Models improve GPT-5.2's performance on RuleArena (Taxes) by 27.4%, reduce Gemini 3 Pro's steps taken in SWE agent tasks by 24.6%, and outperform static prompt optimizers in personalizing GPT-5 to user preferences (85-100% vs. 40-60%). We also find that advi"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2510.02453","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/2510.02453/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":"2510.02453","created_at":"2026-05-20T00:01:35.832908+00:00"},{"alias_kind":"arxiv_version","alias_value":"2510.02453v3","created_at":"2026-05-20T00:01:35.832908+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2510.02453","created_at":"2026-05-20T00:01:35.832908+00:00"},{"alias_kind":"pith_short_12","alias_value":"75OYLG6NPQSB","created_at":"2026-05-20T00:01:35.832908+00:00"},{"alias_kind":"pith_short_16","alias_value":"75OYLG6NPQSBPO2Q","created_at":"2026-05-20T00:01:35.832908+00:00"},{"alias_kind":"pith_short_8","alias_value":"75OYLG6N","created_at":"2026-05-20T00:01:35.832908+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":4,"internal_anchor_count":4,"sample":[{"citing_arxiv_id":"2605.17172","citing_title":"OpenJarvis: Personal AI, On Personal Devices","ref_index":7,"is_internal_anchor":true},{"citing_arxiv_id":"2605.10913","citing_title":"Shepherd: A Runtime Substrate Empowering Meta-Agents with a Formalized Execution Trace","ref_index":3,"is_internal_anchor":true},{"citing_arxiv_id":"2604.09741","citing_title":"ExecTune: Effective Steering of Black-Box LLMs with Guide Models","ref_index":4,"is_internal_anchor":true},{"citing_arxiv_id":"2605.07039","citing_title":"PACEvolve++: Improving Test-time Learning for Evolutionary Search Agents","ref_index":3,"is_internal_anchor":true}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/75OYLG6NPQSBPO2QLDEYQ5ATNU","json":"https://pith.science/pith/75OYLG6NPQSBPO2QLDEYQ5ATNU.json","graph_json":"https://pith.science/api/pith-number/75OYLG6NPQSBPO2QLDEYQ5ATNU/graph.json","events_json":"https://pith.science/api/pith-number/75OYLG6NPQSBPO2QLDEYQ5ATNU/events.json","paper":"https://pith.science/paper/75OYLG6N"},"agent_actions":{"view_html":"https://pith.science/pith/75OYLG6NPQSBPO2QLDEYQ5ATNU","download_json":"https://pith.science/pith/75OYLG6NPQSBPO2QLDEYQ5ATNU.json","view_paper":"https://pith.science/paper/75OYLG6N","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2510.02453&json=true","fetch_graph":"https://pith.science/api/pith-number/75OYLG6NPQSBPO2QLDEYQ5ATNU/graph.json","fetch_events":"https://pith.science/api/pith-number/75OYLG6NPQSBPO2QLDEYQ5ATNU/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/75OYLG6NPQSBPO2QLDEYQ5ATNU/action/timestamp_anchor","attest_storage":"https://pith.science/pith/75OYLG6NPQSBPO2QLDEYQ5ATNU/action/storage_attestation","attest_author":"https://pith.science/pith/75OYLG6NPQSBPO2QLDEYQ5ATNU/action/author_attestation","sign_citation":"https://pith.science/pith/75OYLG6NPQSBPO2QLDEYQ5ATNU/action/citation_signature","submit_replication":"https://pith.science/pith/75OYLG6NPQSBPO2QLDEYQ5ATNU/action/replication_record"}},"created_at":"2026-05-20T00:01:35.832908+00:00","updated_at":"2026-05-20T00:01:35.832908+00:00"}