{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2023:YXMEXGXS52RTL2O3ANDRCXKMYV","short_pith_number":"pith:YXMEXGXS","schema_version":"1.0","canonical_sha256":"c5d84b9af2eea335e9db0347115d4cc57b95e0d6cf1c8ea4ce95ca2b4887dbd5","source":{"kind":"arxiv","id":"2311.11202","version":2},"attestation_state":"computed","paper":{"title":"Unmasking and Improving Data Credibility: A Study with Datasets for Training Harmless Language Models","license":"http://creativecommons.org/licenses/by-nc-nd/4.0/","headline":"","cross_cats":["cs.AI","cs.CL","cs.CY"],"primary_cat":"cs.LG","authors_text":"Hao Cheng, Jialu Wang, Yang Liu, Zhaowei Zhu","submitted_at":"2023-11-19T02:34:12Z","abstract_excerpt":"Language models have shown promise in various tasks but can be affected by undesired data during training, fine-tuning, or alignment. For example, if some unsafe conversations are wrongly annotated as safe ones, the model fine-tuned on these samples may be harmful. Therefore, the correctness of annotations, i.e., the credibility of the dataset, is important. This study focuses on the credibility of real-world datasets, including the popular benchmarks Jigsaw Civil Comments, Anthropic Harmless & Red Team, PKU BeaverTails & SafeRLHF, that can be used for training a harmless language model. Given"},"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":"2311.11202","kind":"arxiv","version":2},"metadata":{"license":"http://creativecommons.org/licenses/by-nc-nd/4.0/","primary_cat":"cs.LG","submitted_at":"2023-11-19T02:34:12Z","cross_cats_sorted":["cs.AI","cs.CL","cs.CY"],"title_canon_sha256":"770af4933170c715951f3c1553d8b7483218c583e0acdb9fd0fed96a6d68e247","abstract_canon_sha256":"5d0ca1e92368399fe3a3a9db50d19167579f08595530f4cb13d0a0f972a7f710"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T07:59:57.440790Z","signature_b64":"4RUxwEaNT3uAwuNmPAshM160AO2dPl/ioe1vPnYIOILoEos53obxxYFC1ym+dIsI6g38iq6przG/f13S8hJaCg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"c5d84b9af2eea335e9db0347115d4cc57b95e0d6cf1c8ea4ce95ca2b4887dbd5","last_reissued_at":"2026-07-05T07:59:57.440330Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T07:59:57.440330Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Unmasking and Improving Data Credibility: A Study with Datasets for Training Harmless Language Models","license":"http://creativecommons.org/licenses/by-nc-nd/4.0/","headline":"","cross_cats":["cs.AI","cs.CL","cs.CY"],"primary_cat":"cs.LG","authors_text":"Hao Cheng, Jialu Wang, Yang Liu, Zhaowei Zhu","submitted_at":"2023-11-19T02:34:12Z","abstract_excerpt":"Language models have shown promise in various tasks but can be affected by undesired data during training, fine-tuning, or alignment. For example, if some unsafe conversations are wrongly annotated as safe ones, the model fine-tuned on these samples may be harmful. Therefore, the correctness of annotations, i.e., the credibility of the dataset, is important. This study focuses on the credibility of real-world datasets, including the popular benchmarks Jigsaw Civil Comments, Anthropic Harmless & Red Team, PKU BeaverTails & SafeRLHF, that can be used for training a harmless language model. Given"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2311.11202","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/2311.11202/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":"2311.11202","created_at":"2026-07-05T07:59:57.440388+00:00"},{"alias_kind":"arxiv_version","alias_value":"2311.11202v2","created_at":"2026-07-05T07:59:57.440388+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2311.11202","created_at":"2026-07-05T07:59:57.440388+00:00"},{"alias_kind":"pith_short_12","alias_value":"YXMEXGXS52RT","created_at":"2026-07-05T07:59:57.440388+00:00"},{"alias_kind":"pith_short_16","alias_value":"YXMEXGXS52RTL2O3","created_at":"2026-07-05T07:59:57.440388+00:00"},{"alias_kind":"pith_short_8","alias_value":"YXMEXGXS","created_at":"2026-07-05T07:59:57.440388+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":4,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.11699","citing_title":"A Data-Centric Framework for Detecting and Correcting Corrupted Labels","ref_index":27,"is_internal_anchor":false},{"citing_arxiv_id":"2606.11695","citing_title":"Noise-Aware Framework for Correcting Corrupted Labels","ref_index":22,"is_internal_anchor":false},{"citing_arxiv_id":"2408.11338","citing_title":"Automatic Dataset Construction (ADC): Sample Collection, Data Curation, and Beyond","ref_index":25,"is_internal_anchor":false},{"citing_arxiv_id":"2604.20544","citing_title":"Evian: Towards Explainable Visual Instruction-tuning Data Auditing","ref_index":9,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/YXMEXGXS52RTL2O3ANDRCXKMYV","json":"https://pith.science/pith/YXMEXGXS52RTL2O3ANDRCXKMYV.json","graph_json":"https://pith.science/api/pith-number/YXMEXGXS52RTL2O3ANDRCXKMYV/graph.json","events_json":"https://pith.science/api/pith-number/YXMEXGXS52RTL2O3ANDRCXKMYV/events.json","paper":"https://pith.science/paper/YXMEXGXS"},"agent_actions":{"view_html":"https://pith.science/pith/YXMEXGXS52RTL2O3ANDRCXKMYV","download_json":"https://pith.science/pith/YXMEXGXS52RTL2O3ANDRCXKMYV.json","view_paper":"https://pith.science/paper/YXMEXGXS","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2311.11202&json=true","fetch_graph":"https://pith.science/api/pith-number/YXMEXGXS52RTL2O3ANDRCXKMYV/graph.json","fetch_events":"https://pith.science/api/pith-number/YXMEXGXS52RTL2O3ANDRCXKMYV/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/YXMEXGXS52RTL2O3ANDRCXKMYV/action/timestamp_anchor","attest_storage":"https://pith.science/pith/YXMEXGXS52RTL2O3ANDRCXKMYV/action/storage_attestation","attest_author":"https://pith.science/pith/YXMEXGXS52RTL2O3ANDRCXKMYV/action/author_attestation","sign_citation":"https://pith.science/pith/YXMEXGXS52RTL2O3ANDRCXKMYV/action/citation_signature","submit_replication":"https://pith.science/pith/YXMEXGXS52RTL2O3ANDRCXKMYV/action/replication_record"}},"created_at":"2026-07-05T07:59:57.440388+00:00","updated_at":"2026-07-05T07:59:57.440388+00:00"}