{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2022:OD4YYGC7KGUTDJW25FWM2UAWEQ","short_pith_number":"pith:OD4YYGC7","schema_version":"1.0","canonical_sha256":"70f98c185f51a931a6dae96ccd50162428e53f975be4d46dd8284835b71316e3","source":{"kind":"arxiv","id":"2205.01663","version":5},"attestation_state":"computed","paper":{"title":"Adversarial Training for High-Stakes Reliability","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI","cs.CL"],"primary_cat":"cs.LG","authors_text":"Adam Scherlis, Ben Weinstein-Raun, Buck Shlegeris, Daniel de Haas, Daniel M. Ziegler, Lawrence Chan, Nate Thomas, Noa Nabeshima, Peter Schmidt-Nielsen, Seraphina Nix, Tao Lin, Tim Bauman","submitted_at":"2022-05-03T17:50:06Z","abstract_excerpt":"In the future, powerful AI systems may be deployed in high-stakes settings, where a single failure could be catastrophic. One technique for improving AI safety in high-stakes settings is adversarial training, which uses an adversary to generate examples to train on in order to achieve better worst-case performance.\n  In this work, we used a safe language generation task (``avoid injuries'') as a testbed for achieving high reliability through adversarial training. We created a series of adversarial training techniques -- including a tool that assists human adversaries -- to find and eliminate f"},"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":"2205.01663","kind":"arxiv","version":5},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2022-05-03T17:50:06Z","cross_cats_sorted":["cs.AI","cs.CL"],"title_canon_sha256":"74b594cede7482800a579a5bc01ab69e46dcf72aaac86616eb11619d006c3621","abstract_canon_sha256":"77771bb98fa9a1545bd8d859f11e2e84902ca9b4e4a2fc86e100e1175d34c0cc"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T05:14:57.203293Z","signature_b64":"Z/1dY0PmLOut5vCHsETevcmkdSYsDmH6xwKKGnr2bT5pkD+a3dVHdPDzIoPmbAYO4fNu7r3VnQjfaQkClgSXBA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"70f98c185f51a931a6dae96ccd50162428e53f975be4d46dd8284835b71316e3","last_reissued_at":"2026-07-05T05:14:57.202844Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T05:14:57.202844Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Adversarial Training for High-Stakes Reliability","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI","cs.CL"],"primary_cat":"cs.LG","authors_text":"Adam Scherlis, Ben Weinstein-Raun, Buck Shlegeris, Daniel de Haas, Daniel M. Ziegler, Lawrence Chan, Nate Thomas, Noa Nabeshima, Peter Schmidt-Nielsen, Seraphina Nix, Tao Lin, Tim Bauman","submitted_at":"2022-05-03T17:50:06Z","abstract_excerpt":"In the future, powerful AI systems may be deployed in high-stakes settings, where a single failure could be catastrophic. One technique for improving AI safety in high-stakes settings is adversarial training, which uses an adversary to generate examples to train on in order to achieve better worst-case performance.\n  In this work, we used a safe language generation task (``avoid injuries'') as a testbed for achieving high reliability through adversarial training. We created a series of adversarial training techniques -- including a tool that assists human adversaries -- to find and eliminate f"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2205.01663","kind":"arxiv","version":5},"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/2205.01663/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":"2205.01663","created_at":"2026-07-05T05:14:57.202901+00:00"},{"alias_kind":"arxiv_version","alias_value":"2205.01663v5","created_at":"2026-07-05T05:14:57.202901+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2205.01663","created_at":"2026-07-05T05:14:57.202901+00:00"},{"alias_kind":"pith_short_12","alias_value":"OD4YYGC7KGUT","created_at":"2026-07-05T05:14:57.202901+00:00"},{"alias_kind":"pith_short_16","alias_value":"OD4YYGC7KGUTDJW2","created_at":"2026-07-05T05:14:57.202901+00:00"},{"alias_kind":"pith_short_8","alias_value":"OD4YYGC7","created_at":"2026-07-05T05:14:57.202901+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":2,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2605.12809","citing_title":"Correcting Influence: Unboxing LLM Outputs with Orthogonal Latent Spaces","ref_index":278,"is_internal_anchor":false},{"citing_arxiv_id":"2209.07858","citing_title":"Red Teaming Language Models to Reduce Harms: Methods, Scaling Behaviors, and Lessons Learned","ref_index":64,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/OD4YYGC7KGUTDJW25FWM2UAWEQ","json":"https://pith.science/pith/OD4YYGC7KGUTDJW25FWM2UAWEQ.json","graph_json":"https://pith.science/api/pith-number/OD4YYGC7KGUTDJW25FWM2UAWEQ/graph.json","events_json":"https://pith.science/api/pith-number/OD4YYGC7KGUTDJW25FWM2UAWEQ/events.json","paper":"https://pith.science/paper/OD4YYGC7"},"agent_actions":{"view_html":"https://pith.science/pith/OD4YYGC7KGUTDJW25FWM2UAWEQ","download_json":"https://pith.science/pith/OD4YYGC7KGUTDJW25FWM2UAWEQ.json","view_paper":"https://pith.science/paper/OD4YYGC7","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2205.01663&json=true","fetch_graph":"https://pith.science/api/pith-number/OD4YYGC7KGUTDJW25FWM2UAWEQ/graph.json","fetch_events":"https://pith.science/api/pith-number/OD4YYGC7KGUTDJW25FWM2UAWEQ/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/OD4YYGC7KGUTDJW25FWM2UAWEQ/action/timestamp_anchor","attest_storage":"https://pith.science/pith/OD4YYGC7KGUTDJW25FWM2UAWEQ/action/storage_attestation","attest_author":"https://pith.science/pith/OD4YYGC7KGUTDJW25FWM2UAWEQ/action/author_attestation","sign_citation":"https://pith.science/pith/OD4YYGC7KGUTDJW25FWM2UAWEQ/action/citation_signature","submit_replication":"https://pith.science/pith/OD4YYGC7KGUTDJW25FWM2UAWEQ/action/replication_record"}},"created_at":"2026-07-05T05:14:57.202901+00:00","updated_at":"2026-07-05T05:14:57.202901+00:00"}