{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2024:PM7VT45FNNDORRZV767TTBFDLA","short_pith_number":"pith:PM7VT45F","schema_version":"1.0","canonical_sha256":"7b3f59f3a56b46e8c735ffbf3984a3583e3da850d17c47f1553640d28f8ce87f","source":{"kind":"arxiv","id":"2406.06369","version":4},"attestation_state":"computed","paper":{"title":"Annotation alignment: Comparing LLM and human annotations of conversational safety","license":"http://creativecommons.org/licenses/by-nc-nd/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Emma Pierson, Pang Wei Koh, Rajiv Movva","submitted_at":"2024-06-10T15:30:13Z","abstract_excerpt":"Do LLMs align with human perceptions of safety? We study this question via annotation alignment, the extent to which LLMs and humans agree when annotating the safety of user-chatbot conversations. We leverage the recent DICES dataset (Aroyo et al., 2023), in which 350 conversations are each rated for safety by 112 annotators spanning 10 race-gender groups. GPT-4 achieves a Pearson correlation of $r = 0.59$ with the average annotator rating, \\textit{higher} than the median annotator's correlation with the average ($r=0.51$). We show that larger datasets are needed to resolve whether LLMs exhibi"},"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":"2406.06369","kind":"arxiv","version":4},"metadata":{"license":"http://creativecommons.org/licenses/by-nc-nd/4.0/","primary_cat":"cs.CL","submitted_at":"2024-06-10T15:30:13Z","cross_cats_sorted":[],"title_canon_sha256":"b7f5a1b4a9b79deacc3426269764d2ff9260c9ced93e0b67eba899ab758d7ba2","abstract_canon_sha256":"d696822ce2f2f77c36a91adf726e8d4bb6b6e7bc9eba3bf253b0bece455ec062"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T09:16:49.935373Z","signature_b64":"jO29L8gXOD6EoOr9y4zX9glovFcvj72G9YDxDC/gaaVsKvhOBFfa68KuB5pxtlfBYQacTAY2gA4NGkc3ldxtBw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"7b3f59f3a56b46e8c735ffbf3984a3583e3da850d17c47f1553640d28f8ce87f","last_reissued_at":"2026-07-05T09:16:49.934889Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T09:16:49.934889Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Annotation alignment: Comparing LLM and human annotations of conversational safety","license":"http://creativecommons.org/licenses/by-nc-nd/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Emma Pierson, Pang Wei Koh, Rajiv Movva","submitted_at":"2024-06-10T15:30:13Z","abstract_excerpt":"Do LLMs align with human perceptions of safety? We study this question via annotation alignment, the extent to which LLMs and humans agree when annotating the safety of user-chatbot conversations. We leverage the recent DICES dataset (Aroyo et al., 2023), in which 350 conversations are each rated for safety by 112 annotators spanning 10 race-gender groups. GPT-4 achieves a Pearson correlation of $r = 0.59$ with the average annotator rating, \\textit{higher} than the median annotator's correlation with the average ($r=0.51$). We show that larger datasets are needed to resolve whether LLMs exhibi"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2406.06369","kind":"arxiv","version":4},"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/2406.06369/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":"2406.06369","created_at":"2026-07-05T09:16:49.934939+00:00"},{"alias_kind":"arxiv_version","alias_value":"2406.06369v4","created_at":"2026-07-05T09:16:49.934939+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2406.06369","created_at":"2026-07-05T09:16:49.934939+00:00"},{"alias_kind":"pith_short_12","alias_value":"PM7VT45FNNDO","created_at":"2026-07-05T09:16:49.934939+00:00"},{"alias_kind":"pith_short_16","alias_value":"PM7VT45FNNDORRZV","created_at":"2026-07-05T09:16:49.934939+00:00"},{"alias_kind":"pith_short_8","alias_value":"PM7VT45F","created_at":"2026-07-05T09:16:49.934939+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":6,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2605.28734","citing_title":"Code as a Weapon: A Consensus-Labeled Prompt Bank for Measuring Coding-Model Compliance with Malicious-Code Requests","ref_index":14,"is_internal_anchor":false},{"citing_arxiv_id":"2606.21296","citing_title":"Discriminatory Compliance: How LLMs Answer Queries from Protected Groups","ref_index":82,"is_internal_anchor":false},{"citing_arxiv_id":"2605.20351","citing_title":"Refusal Evaluation in Coding LLMs and Code Agents: A Systematic Review of Thirteen Malicious-Code Prompt Corpora (2023-2025)","ref_index":38,"is_internal_anchor":false},{"citing_arxiv_id":"2507.00439","citing_title":"Improving the Distributional Alignment of LLMs using Supervision","ref_index":23,"is_internal_anchor":false},{"citing_arxiv_id":"2605.05329","citing_title":"Understanding Annotator Safety Policy with Interpretability","ref_index":75,"is_internal_anchor":false},{"citing_arxiv_id":"2605.03179","citing_title":"A Validated Prompt Bank for Malicious Code Generation: Separating Executable Weapons from Security Knowledge in 1,554 Consensus-Labeled Prompts","ref_index":43,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/PM7VT45FNNDORRZV767TTBFDLA","json":"https://pith.science/pith/PM7VT45FNNDORRZV767TTBFDLA.json","graph_json":"https://pith.science/api/pith-number/PM7VT45FNNDORRZV767TTBFDLA/graph.json","events_json":"https://pith.science/api/pith-number/PM7VT45FNNDORRZV767TTBFDLA/events.json","paper":"https://pith.science/paper/PM7VT45F"},"agent_actions":{"view_html":"https://pith.science/pith/PM7VT45FNNDORRZV767TTBFDLA","download_json":"https://pith.science/pith/PM7VT45FNNDORRZV767TTBFDLA.json","view_paper":"https://pith.science/paper/PM7VT45F","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2406.06369&json=true","fetch_graph":"https://pith.science/api/pith-number/PM7VT45FNNDORRZV767TTBFDLA/graph.json","fetch_events":"https://pith.science/api/pith-number/PM7VT45FNNDORRZV767TTBFDLA/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/PM7VT45FNNDORRZV767TTBFDLA/action/timestamp_anchor","attest_storage":"https://pith.science/pith/PM7VT45FNNDORRZV767TTBFDLA/action/storage_attestation","attest_author":"https://pith.science/pith/PM7VT45FNNDORRZV767TTBFDLA/action/author_attestation","sign_citation":"https://pith.science/pith/PM7VT45FNNDORRZV767TTBFDLA/action/citation_signature","submit_replication":"https://pith.science/pith/PM7VT45FNNDORRZV767TTBFDLA/action/replication_record"}},"created_at":"2026-07-05T09:16:49.934939+00:00","updated_at":"2026-07-05T09:16:49.934939+00:00"}