{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:7RKPCNCXEISI65D3XVC75BJ666","short_pith_number":"pith:7RKPCNCX","schema_version":"1.0","canonical_sha256":"fc54f1345722248f747bbd45fe853ef7bc9b73bde95d7566c1bdcbb51c255e02","source":{"kind":"arxiv","id":"2506.01709","version":1},"attestation_state":"computed","paper":{"title":"Fairness Dynamics During Training","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Barry-John Theobald, Krishna Patel, Luca Zappella, Nicholas Apostoloff, Nivedha Sivakumar","submitted_at":"2025-06-02T14:15:37Z","abstract_excerpt":"We investigate fairness dynamics during Large Language Model (LLM) training to enable the diagnoses of biases and mitigations through training interventions like early stopping; we find that biases can emerge suddenly and do not always follow common performance metrics. We introduce two new metrics to evaluate fairness dynamics holistically during model pre-training: Average Rank and Jensen-Shannon Divergence by Parts. These metrics provide insights into the Pythia models' progression of biases in gender prediction of occupations on the WinoBias dataset. By monitoring these dynamics, we find t"},"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":"2506.01709","kind":"arxiv","version":1},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CL","submitted_at":"2025-06-02T14:15:37Z","cross_cats_sorted":[],"title_canon_sha256":"15cd9b92519fa2d717a785ead52cb1f0106793c66ca7dd34b6a1c58f9d2cde83","abstract_canon_sha256":"337a23c6234e5bed3efa72787973e99cc605337fa2afb31e269bed72e71fbe84"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:14:18.145875Z","signature_b64":"neEX8Hxs/WhcXQbkz3IcnKM8h5usyB0OZpLGVGYnNnFTeAcATG4wdpYlUieD+J+sgYgERVpUPw1hR/L/AnFrBw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"fc54f1345722248f747bbd45fe853ef7bc9b73bde95d7566c1bdcbb51c255e02","last_reissued_at":"2026-07-05T11:14:18.145403Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:14:18.145403Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Fairness Dynamics During Training","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Barry-John Theobald, Krishna Patel, Luca Zappella, Nicholas Apostoloff, Nivedha Sivakumar","submitted_at":"2025-06-02T14:15:37Z","abstract_excerpt":"We investigate fairness dynamics during Large Language Model (LLM) training to enable the diagnoses of biases and mitigations through training interventions like early stopping; we find that biases can emerge suddenly and do not always follow common performance metrics. We introduce two new metrics to evaluate fairness dynamics holistically during model pre-training: Average Rank and Jensen-Shannon Divergence by Parts. These metrics provide insights into the Pythia models' progression of biases in gender prediction of occupations on the WinoBias dataset. By monitoring these dynamics, we find t"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2506.01709","kind":"arxiv","version":1},"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/2506.01709/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":"2506.01709","created_at":"2026-07-05T11:14:18.145459+00:00"},{"alias_kind":"arxiv_version","alias_value":"2506.01709v1","created_at":"2026-07-05T11:14:18.145459+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2506.01709","created_at":"2026-07-05T11:14:18.145459+00:00"},{"alias_kind":"pith_short_12","alias_value":"7RKPCNCXEISI","created_at":"2026-07-05T11:14:18.145459+00:00"},{"alias_kind":"pith_short_16","alias_value":"7RKPCNCXEISI65D3","created_at":"2026-07-05T11:14:18.145459+00:00"},{"alias_kind":"pith_short_8","alias_value":"7RKPCNCX","created_at":"2026-07-05T11:14:18.145459+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":0,"internal_anchor_count":0,"sample":[]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/7RKPCNCXEISI65D3XVC75BJ666","json":"https://pith.science/pith/7RKPCNCXEISI65D3XVC75BJ666.json","graph_json":"https://pith.science/api/pith-number/7RKPCNCXEISI65D3XVC75BJ666/graph.json","events_json":"https://pith.science/api/pith-number/7RKPCNCXEISI65D3XVC75BJ666/events.json","paper":"https://pith.science/paper/7RKPCNCX"},"agent_actions":{"view_html":"https://pith.science/pith/7RKPCNCXEISI65D3XVC75BJ666","download_json":"https://pith.science/pith/7RKPCNCXEISI65D3XVC75BJ666.json","view_paper":"https://pith.science/paper/7RKPCNCX","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2506.01709&json=true","fetch_graph":"https://pith.science/api/pith-number/7RKPCNCXEISI65D3XVC75BJ666/graph.json","fetch_events":"https://pith.science/api/pith-number/7RKPCNCXEISI65D3XVC75BJ666/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/7RKPCNCXEISI65D3XVC75BJ666/action/timestamp_anchor","attest_storage":"https://pith.science/pith/7RKPCNCXEISI65D3XVC75BJ666/action/storage_attestation","attest_author":"https://pith.science/pith/7RKPCNCXEISI65D3XVC75BJ666/action/author_attestation","sign_citation":"https://pith.science/pith/7RKPCNCXEISI65D3XVC75BJ666/action/citation_signature","submit_replication":"https://pith.science/pith/7RKPCNCXEISI65D3XVC75BJ666/action/replication_record"}},"created_at":"2026-07-05T11:14:18.145459+00:00","updated_at":"2026-07-05T11:14:18.145459+00:00"}