{"bundle_type":"pith_open_graph_bundle","bundle_version":"1.0","pith_number":"pith:2021:URRXZ7SKKAT4VJ3QNXJBHGX4EV","short_pith_number":"pith:URRXZ7SK","canonical_record":{"source":{"id":"2108.07790","kind":"arxiv","version":3},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CL","submitted_at":"2021-08-04T22:18:10Z","cross_cats_sorted":["cs.LG"],"title_canon_sha256":"37a9fd7aa35f6d2e5854e35666c7566ef85b47c78478c7e4ee4e214fddfd39d0","abstract_canon_sha256":"1fd99f798d5e3e1ea8271fe71cda43a44feeca748123c78c3a1324fcd6b44a5e"},"schema_version":"1.0"},"canonical_sha256":"a4637cfe4a5027caa7706dd2139afc255b9c76730f2285b57823f7f3581c3db9","source":{"kind":"arxiv","id":"2108.07790","version":3},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2108.07790","created_at":"2026-07-05T03:35:31Z"},{"alias_kind":"arxiv_version","alias_value":"2108.07790v3","created_at":"2026-07-05T03:35:31Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2108.07790","created_at":"2026-07-05T03:35:31Z"},{"alias_kind":"pith_short_12","alias_value":"URRXZ7SKKAT4","created_at":"2026-07-05T03:35:31Z"},{"alias_kind":"pith_short_16","alias_value":"URRXZ7SKKAT4VJ3Q","created_at":"2026-07-05T03:35:31Z"},{"alias_kind":"pith_short_8","alias_value":"URRXZ7SK","created_at":"2026-07-05T03:35:31Z"}],"events":[{"event_type":"record_created","subject_pith_number":"pith:2021:URRXZ7SKKAT4VJ3QNXJBHGX4EV","target":"record","payload":{"canonical_record":{"source":{"id":"2108.07790","kind":"arxiv","version":3},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CL","submitted_at":"2021-08-04T22:18:10Z","cross_cats_sorted":["cs.LG"],"title_canon_sha256":"37a9fd7aa35f6d2e5854e35666c7566ef85b47c78478c7e4ee4e214fddfd39d0","abstract_canon_sha256":"1fd99f798d5e3e1ea8271fe71cda43a44feeca748123c78c3a1324fcd6b44a5e"},"schema_version":"1.0"},"canonical_sha256":"a4637cfe4a5027caa7706dd2139afc255b9c76730f2285b57823f7f3581c3db9","receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T03:35:31.746940Z","signature_b64":"FrXvke+TJbNN9WoIYSFhB8KyIsVkQ6tTEC99c991MKsxdg/M4QX3EsCzE79Uf+zTU++0amfmUVA/UARwPlpvBg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"a4637cfe4a5027caa7706dd2139afc255b9c76730f2285b57823f7f3581c3db9","last_reissued_at":"2026-07-05T03:35:31.746474Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T03:35:31.746474Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"source_kind":"arxiv","source_id":"2108.07790","source_version":3,"attestation_state":"computed"},"signer":{"signer_id":"pith.science","signer_type":"pith_registry","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"created_at":"2026-07-05T03:35:31Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"oXv8lQCOkTf4+3VByzu2SOjkHth43yY3IqnhvykF1apG0BXoeI3jzvHMYEeNBBt6PDB2QeiM7BtBRt6biLHlBg==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-07-31T19:40:18.866956Z"},"content_sha256":"b825e53aec7885e0752b12d263ea83f887636fcb592b5bd9fceda25f5e42d647","schema_version":"1.0","event_id":"sha256:b825e53aec7885e0752b12d263ea83f887636fcb592b5bd9fceda25f5e42d647"},{"event_type":"graph_snapshot","subject_pith_number":"pith:2021:URRXZ7SKKAT4VJ3QNXJBHGX4EV","target":"graph","payload":{"graph_snapshot":{"paper":{"title":"Mitigating harm in language models with conditional-likelihood filtration","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.LG"],"primary_cat":"cs.CL","authors_text":"Adrien Morisot, Carol Chen, Cooper Raterink, Helen Ngo, Ivan Zhang, Jo\\~ao G.M. Ara\\'ujo, Nicholas Frosst","submitted_at":"2021-08-04T22:18:10Z","abstract_excerpt":"Language models trained on large-scale unfiltered datasets curated from the open web acquire systemic biases, prejudices, and harmful views from their training data. We present a methodology for programmatically identifying and removing harmful text from web-scale datasets. A pretrained language model is used to calculate the log-likelihood of researcher-written trigger phrases conditioned on a specific document, which is used to identify and filter documents from the dataset. We demonstrate that models trained on this filtered dataset exhibit lower propensity to generate harmful text, with a "},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2108.07790","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/2108.07790/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":"2026-07-05T03:35:31Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"luUHiAS4BOsqWM0+8I9/+Po0kD2q4BsEQE+rXH4sL/mMrrlQv32HF5C/pAh/eJSkNRMlxPc8aUWGfmoA612vCQ==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-07-31T19:40:18.867487Z"},"content_sha256":"4b0b5e517ea65b63ae951da384ea0d7bfb6ac3ae39bea624fd6b1c8f5fd4dd2e","schema_version":"1.0","event_id":"sha256:4b0b5e517ea65b63ae951da384ea0d7bfb6ac3ae39bea624fd6b1c8f5fd4dd2e"}],"timestamp_proofs":[],"mirror_hints":[{"mirror_type":"https","name":"Pith Resolver","base_url":"https://pith.science","bundle_url":"https://pith.science/pith/URRXZ7SKKAT4VJ3QNXJBHGX4EV/bundle.json","state_url":"https://pith.science/pith/URRXZ7SKKAT4VJ3QNXJBHGX4EV/state.json","well_known_bundle_url":"https://pith.science/.well-known/pith/URRXZ7SKKAT4VJ3QNXJBHGX4EV/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":"2026-07-31T19:40:18Z","links":{"resolver":"https://pith.science/pith/URRXZ7SKKAT4VJ3QNXJBHGX4EV","bundle":"https://pith.science/pith/URRXZ7SKKAT4VJ3QNXJBHGX4EV/bundle.json","state":"https://pith.science/pith/URRXZ7SKKAT4VJ3QNXJBHGX4EV/state.json","well_known_bundle":"https://pith.science/.well-known/pith/URRXZ7SKKAT4VJ3QNXJBHGX4EV/bundle.json"},"state":{"state_type":"pith_open_graph_state","state_version":"1.0","pith_number":"pith:2021:URRXZ7SKKAT4VJ3QNXJBHGX4EV","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":"1fd99f798d5e3e1ea8271fe71cda43a44feeca748123c78c3a1324fcd6b44a5e","cross_cats_sorted":["cs.LG"],"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CL","submitted_at":"2021-08-04T22:18:10Z","title_canon_sha256":"37a9fd7aa35f6d2e5854e35666c7566ef85b47c78478c7e4ee4e214fddfd39d0"},"schema_version":"1.0","source":{"id":"2108.07790","kind":"arxiv","version":3}},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2108.07790","created_at":"2026-07-05T03:35:31Z"},{"alias_kind":"arxiv_version","alias_value":"2108.07790v3","created_at":"2026-07-05T03:35:31Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2108.07790","created_at":"2026-07-05T03:35:31Z"},{"alias_kind":"pith_short_12","alias_value":"URRXZ7SKKAT4","created_at":"2026-07-05T03:35:31Z"},{"alias_kind":"pith_short_16","alias_value":"URRXZ7SKKAT4VJ3Q","created_at":"2026-07-05T03:35:31Z"},{"alias_kind":"pith_short_8","alias_value":"URRXZ7SK","created_at":"2026-07-05T03:35:31Z"}],"graph_snapshots":[{"event_id":"sha256:4b0b5e517ea65b63ae951da384ea0d7bfb6ac3ae39bea624fd6b1c8f5fd4dd2e","target":"graph","created_at":"2026-07-05T03:35:31Z","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/2108.07790/integrity.json","findings":[],"snapshot_sha256":"c28c3603d3b5d939e8dc4c7e95fa8dfce3d595e45f758748cecf8e644a296938","summary":{"advisory":0,"by_detector":{},"critical":0,"informational":0}},"paper":{"abstract_excerpt":"Language models trained on large-scale unfiltered datasets curated from the open web acquire systemic biases, prejudices, and harmful views from their training data. We present a methodology for programmatically identifying and removing harmful text from web-scale datasets. A pretrained language model is used to calculate the log-likelihood of researcher-written trigger phrases conditioned on a specific document, which is used to identify and filter documents from the dataset. We demonstrate that models trained on this filtered dataset exhibit lower propensity to generate harmful text, with a ","authors_text":"Adrien Morisot, Carol Chen, Cooper Raterink, Helen Ngo, Ivan Zhang, Jo\\~ao G.M. Ara\\'ujo, Nicholas Frosst","cross_cats":["cs.LG"],"headline":"","license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CL","submitted_at":"2021-08-04T22:18:10Z","title":"Mitigating harm in language models with conditional-likelihood filtration"},"references":{"count":0,"internal_anchors":0,"resolved_work":0,"sample":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2108.07790","kind":"arxiv","version":3},"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:b825e53aec7885e0752b12d263ea83f887636fcb592b5bd9fceda25f5e42d647","target":"record","created_at":"2026-07-05T03:35:31Z","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":"1fd99f798d5e3e1ea8271fe71cda43a44feeca748123c78c3a1324fcd6b44a5e","cross_cats_sorted":["cs.LG"],"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CL","submitted_at":"2021-08-04T22:18:10Z","title_canon_sha256":"37a9fd7aa35f6d2e5854e35666c7566ef85b47c78478c7e4ee4e214fddfd39d0"},"schema_version":"1.0","source":{"id":"2108.07790","kind":"arxiv","version":3}},"canonical_sha256":"a4637cfe4a5027caa7706dd2139afc255b9c76730f2285b57823f7f3581c3db9","receipt":{"algorithm":"ed25519","builder_version":"pith-number-builder-2026-05-17-v1","canonical_sha256":"a4637cfe4a5027caa7706dd2139afc255b9c76730f2285b57823f7f3581c3db9","first_computed_at":"2026-07-05T03:35:31.746474Z","key_id":"pith-v1-2026-05","kind":"pith_receipt","last_reissued_at":"2026-07-05T03:35:31.746474Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","receipt_version":"0.3","signature_b64":"FrXvke+TJbNN9WoIYSFhB8KyIsVkQ6tTEC99c991MKsxdg/M4QX3EsCzE79Uf+zTU++0amfmUVA/UARwPlpvBg==","signature_status":"signed_v1","signed_at":"2026-07-05T03:35:31.746940Z","signed_message":"canonical_sha256_bytes"},"source_id":"2108.07790","source_kind":"arxiv","source_version":3}}},"equivocations":[],"invalid_events":[],"applied_event_ids":["sha256:b825e53aec7885e0752b12d263ea83f887636fcb592b5bd9fceda25f5e42d647","sha256:4b0b5e517ea65b63ae951da384ea0d7bfb6ac3ae39bea624fd6b1c8f5fd4dd2e"],"state_sha256":"818d5c0d8ba2ee482fb8cdc18dccddfee857efd241c73670acc496c9833180b4"},"bundle_signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"6qpJrTAhuxy6Ha8TuUIDmFC0OKPx6yKxi0El62qhfB/qB1H01vR00q5E+ZUezr7qBpYqbWdKJg1aMvZcBNrACg==","signed_message":"bundle_sha256_bytes","signed_at":"2026-07-31T19:40:18.873171Z","bundle_sha256":"bf17897109a44ae4631b2a7396ee18bf0ffe4c3e45be30633d34c543f6f50e62"}}