{"bundle_type":"pith_open_graph_bundle","bundle_version":"1.0","pith_number":"pith:2023:OMFPDWZCPAFNZWGZPYYIW26ROI","short_pith_number":"pith:OMFPDWZC","canonical_record":{"source":{"id":"2306.07567","kind":"arxiv","version":2},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2023-06-13T06:40:37Z","cross_cats_sorted":["cs.CL"],"title_canon_sha256":"abd386c93f78630d097902ddc6aff4ab4a15ae6eeddf4f5b554a4e4d74f6b062","abstract_canon_sha256":"5b2e8cbe4a24c9f97848c377755553e76933808c2949f5dd4b1cc3f05f2ca112"},"schema_version":"1.0"},"canonical_sha256":"730af1db22780adcd8d97e308b6bd1723da196414b2ce768a55cdcbda7a93256","source":{"kind":"arxiv","id":"2306.07567","version":2},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2306.07567","created_at":"2026-07-05T06:21:17Z"},{"alias_kind":"arxiv_version","alias_value":"2306.07567v2","created_at":"2026-07-05T06:21:17Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2306.07567","created_at":"2026-07-05T06:21:17Z"},{"alias_kind":"pith_short_12","alias_value":"OMFPDWZCPAFN","created_at":"2026-07-05T06:21:17Z"},{"alias_kind":"pith_short_16","alias_value":"OMFPDWZCPAFNZWGZ","created_at":"2026-07-05T06:21:17Z"},{"alias_kind":"pith_short_8","alias_value":"OMFPDWZC","created_at":"2026-07-05T06:21:17Z"}],"events":[{"event_type":"record_created","subject_pith_number":"pith:2023:OMFPDWZCPAFNZWGZPYYIW26ROI","target":"record","payload":{"canonical_record":{"source":{"id":"2306.07567","kind":"arxiv","version":2},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2023-06-13T06:40:37Z","cross_cats_sorted":["cs.CL"],"title_canon_sha256":"abd386c93f78630d097902ddc6aff4ab4a15ae6eeddf4f5b554a4e4d74f6b062","abstract_canon_sha256":"5b2e8cbe4a24c9f97848c377755553e76933808c2949f5dd4b1cc3f05f2ca112"},"schema_version":"1.0"},"canonical_sha256":"730af1db22780adcd8d97e308b6bd1723da196414b2ce768a55cdcbda7a93256","receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T06:21:17.969877Z","signature_b64":"ZQM769xNkzytILqK7av3Yn1/sW8GhLd0/DDmgGU7xlaHAVLZQ5MhsMISWXNBxIRKD5yUjKk3jhfw9g+alZ85AQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"730af1db22780adcd8d97e308b6bd1723da196414b2ce768a55cdcbda7a93256","last_reissued_at":"2026-07-05T06:21:17.969402Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T06:21:17.969402Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"source_kind":"arxiv","source_id":"2306.07567","source_version":2,"attestation_state":"computed"},"signer":{"signer_id":"pith.science","signer_type":"pith_registry","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"created_at":"2026-07-05T06:21:17Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"lSB7n34oB8YZWKeI3cu/h2si7SAZ0m2eyGPJ0j9AAgQIeBUhwWNjBojkMy1V0zAE3lC9O4PiPJYfjSdyd1SBDw==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-08-05T04:54:07.445417Z"},"content_sha256":"63f697029181f1563eb94bfbc12eb04ac34ca5e969cbde6598a49de3bc5630e3","schema_version":"1.0","event_id":"sha256:63f697029181f1563eb94bfbc12eb04ac34ca5e969cbde6598a49de3bc5630e3"},{"event_type":"graph_snapshot","subject_pith_number":"pith:2023:OMFPDWZCPAFNZWGZPYYIW26ROI","target":"graph","payload":{"graph_snapshot":{"paper":{"title":"Large Language Models Sometimes Generate Purely Negatively-Reinforced Text","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.CL"],"primary_cat":"cs.LG","authors_text":"Fabien Roger","submitted_at":"2023-06-13T06:40:37Z","abstract_excerpt":"When using adversarial training, it is common practice to train against the most egregious failures. However, this might imply using examples with sensitive information (such as leaked passwords or security vulnerabilities) as training data. One might assume that language models trained with gradient descent never generate text snippets which were only present in examples associated with the lowest possible reward. In this paper, we show that this assumption is wrong: in some situations, large language models do learn from such negatively-reinforced examples. We present a specific training set"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2306.07567","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/2306.07567/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"},"verdict_id":null},"signer":{"signer_id":"pith.science","signer_type":"pith_registry","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"created_at":"2026-07-05T06:21:17Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"IG49IfJosc0QOo+04A9r2ImJY7HtMULt2rxUzQgLWhtb9IBiMJ6hpYF9ki6hV3qmEX0LvXg/Wl3NjIxF0rHeAg==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-08-05T04:54:07.445908Z"},"content_sha256":"d2f630a410e0a42e6d3131bbece225a1dd2f449a29c25c34c2537b1de501e8b5","schema_version":"1.0","event_id":"sha256:d2f630a410e0a42e6d3131bbece225a1dd2f449a29c25c34c2537b1de501e8b5"}],"timestamp_proofs":[],"mirror_hints":[{"mirror_type":"https","name":"Pith Resolver","base_url":"https://pith.science","bundle_url":"https://pith.science/pith/OMFPDWZCPAFNZWGZPYYIW26ROI/bundle.json","state_url":"https://pith.science/pith/OMFPDWZCPAFNZWGZPYYIW26ROI/state.json","well_known_bundle_url":"https://pith.science/.well-known/pith/OMFPDWZCPAFNZWGZPYYIW26ROI/bundle.json","status":"primary"}],"public_keys":[{"key_id":"pith-v1-2026-05","algorithm":"ed25519","format":"raw","public_key_b64":"stVStoiQhXFxp4s2pdzPNoqVNBMojDU/fJ2db5S3CbM=","public_key_hex":"b2d552b68890857171a78b36a5dccf368a953413288c353f7c9d9d6f94b709b3","fingerprint_sha256_b32_first128bits":"RVFV5Z2OI2J3ZUO7ERDEBCYNKS","fingerprint_sha256_hex":"8d4b5ee74e4693bcd1df2446408b0d54","rotates_at":null,"url":"https://pith.science/pith-signing-key.json","notes":"Pith uses this Ed25519 key to sign canonical record SHA-256 digests. Verify with: ed25519_verify(public_key, message=canonical_sha256_bytes, signature=base64decode(signature_b64))."}],"merge_version":"pith-open-graph-merge-v1","built_at":"2026-08-05T04:54:07Z","links":{"resolver":"https://pith.science/pith/OMFPDWZCPAFNZWGZPYYIW26ROI","bundle":"https://pith.science/pith/OMFPDWZCPAFNZWGZPYYIW26ROI/bundle.json","state":"https://pith.science/pith/OMFPDWZCPAFNZWGZPYYIW26ROI/state.json","well_known_bundle":"https://pith.science/.well-known/pith/OMFPDWZCPAFNZWGZPYYIW26ROI/bundle.json"},"state":{"state_type":"pith_open_graph_state","state_version":"1.0","pith_number":"pith:2023:OMFPDWZCPAFNZWGZPYYIW26ROI","merge_version":"pith-open-graph-merge-v1","event_count":2,"valid_event_count":2,"invalid_event_count":0,"equivocation_count":0,"current":{"canonical_record":{"metadata":{"abstract_canon_sha256":"5b2e8cbe4a24c9f97848c377755553e76933808c2949f5dd4b1cc3f05f2ca112","cross_cats_sorted":["cs.CL"],"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2023-06-13T06:40:37Z","title_canon_sha256":"abd386c93f78630d097902ddc6aff4ab4a15ae6eeddf4f5b554a4e4d74f6b062"},"schema_version":"1.0","source":{"id":"2306.07567","kind":"arxiv","version":2}},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2306.07567","created_at":"2026-07-05T06:21:17Z"},{"alias_kind":"arxiv_version","alias_value":"2306.07567v2","created_at":"2026-07-05T06:21:17Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2306.07567","created_at":"2026-07-05T06:21:17Z"},{"alias_kind":"pith_short_12","alias_value":"OMFPDWZCPAFN","created_at":"2026-07-05T06:21:17Z"},{"alias_kind":"pith_short_16","alias_value":"OMFPDWZCPAFNZWGZ","created_at":"2026-07-05T06:21:17Z"},{"alias_kind":"pith_short_8","alias_value":"OMFPDWZC","created_at":"2026-07-05T06:21:17Z"}],"graph_snapshots":[{"event_id":"sha256:d2f630a410e0a42e6d3131bbece225a1dd2f449a29c25c34c2537b1de501e8b5","target":"graph","created_at":"2026-07-05T06:21:17Z","signer":{"key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signer_id":"pith.science","signer_type":"pith_registry"},"payload":{"graph_snapshot":{"author_claims":{"count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57","strong_count":0},"builder_version":"pith-number-builder-2026-05-17-v1","claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"formal_canon":{"evidence_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"integrity":{"available":true,"clean":true,"detectors_run":[],"endpoint":"/pith/2306.07567/integrity.json","findings":[],"snapshot_sha256":"c28c3603d3b5d939e8dc4c7e95fa8dfce3d595e45f758748cecf8e644a296938","summary":{"advisory":0,"by_detector":{},"critical":0,"informational":0}},"paper":{"abstract_excerpt":"When using adversarial training, it is common practice to train against the most egregious failures. However, this might imply using examples with sensitive information (such as leaked passwords or security vulnerabilities) as training data. One might assume that language models trained with gradient descent never generate text snippets which were only present in examples associated with the lowest possible reward. In this paper, we show that this assumption is wrong: in some situations, large language models do learn from such negatively-reinforced examples. We present a specific training set","authors_text":"Fabien Roger","cross_cats":["cs.CL"],"headline":"","license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2023-06-13T06:40:37Z","title":"Large Language Models Sometimes Generate Purely Negatively-Reinforced Text"},"references":{"count":0,"internal_anchors":0,"resolved_work":0,"sample":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2306.07567","kind":"arxiv","version":2},"verdict":{"created_at":null,"id":null,"model_set":{},"one_line_summary":"","pipeline_version":null,"pith_extraction_headline":"","strongest_claim":"","weakest_assumption":""}},"verdict_id":null}}],"author_attestations":[],"timestamp_anchors":[],"storage_attestations":[],"citation_signatures":[],"replication_records":[],"corrections":[],"mirror_hints":[],"record_created":{"event_id":"sha256:63f697029181f1563eb94bfbc12eb04ac34ca5e969cbde6598a49de3bc5630e3","target":"record","created_at":"2026-07-05T06:21:17Z","signer":{"key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signer_id":"pith.science","signer_type":"pith_registry"},"payload":{"attestation_state":"computed","canonical_record":{"metadata":{"abstract_canon_sha256":"5b2e8cbe4a24c9f97848c377755553e76933808c2949f5dd4b1cc3f05f2ca112","cross_cats_sorted":["cs.CL"],"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2023-06-13T06:40:37Z","title_canon_sha256":"abd386c93f78630d097902ddc6aff4ab4a15ae6eeddf4f5b554a4e4d74f6b062"},"schema_version":"1.0","source":{"id":"2306.07567","kind":"arxiv","version":2}},"canonical_sha256":"730af1db22780adcd8d97e308b6bd1723da196414b2ce768a55cdcbda7a93256","receipt":{"algorithm":"ed25519","builder_version":"pith-number-builder-2026-05-17-v1","canonical_sha256":"730af1db22780adcd8d97e308b6bd1723da196414b2ce768a55cdcbda7a93256","first_computed_at":"2026-07-05T06:21:17.969402Z","key_id":"pith-v1-2026-05","kind":"pith_receipt","last_reissued_at":"2026-07-05T06:21:17.969402Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","receipt_version":"0.3","signature_b64":"ZQM769xNkzytILqK7av3Yn1/sW8GhLd0/DDmgGU7xlaHAVLZQ5MhsMISWXNBxIRKD5yUjKk3jhfw9g+alZ85AQ==","signature_status":"signed_v1","signed_at":"2026-07-05T06:21:17.969877Z","signed_message":"canonical_sha256_bytes"},"source_id":"2306.07567","source_kind":"arxiv","source_version":2}}},"equivocations":[],"invalid_events":[],"applied_event_ids":["sha256:63f697029181f1563eb94bfbc12eb04ac34ca5e969cbde6598a49de3bc5630e3","sha256:d2f630a410e0a42e6d3131bbece225a1dd2f449a29c25c34c2537b1de501e8b5"],"state_sha256":"3749398cd230720d064fe4a4963a7929d7ed63ab11294e9c02b9b0a879a7c2a8"},"bundle_signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"GOAiGkwg1CDCXIChjI4gHrKBjF3sDs+vP6pD4KkQu8aXLgIvkcwnml2cORqxUD2O7d0GysPHJzOar2jmHULpBQ==","signed_message":"bundle_sha256_bytes","signed_at":"2026-08-05T04:54:07.449506Z","bundle_sha256":"100b42489469592602e686692cac846e973dfd538871b872c77c4be42848a00e"}}