{"bundle_type":"pith_open_graph_bundle","bundle_version":"1.0","pith_number":"pith:2026:L3CZPBVBFUOT4DLWAVW47Q4FIY","short_pith_number":"pith:L3CZPBVB","canonical_record":{"source":{"id":"2606.04807","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.AI","submitted_at":"2026-06-03T12:31:42Z","cross_cats_sorted":["cs.CL","cs.CY","cs.LG"],"title_canon_sha256":"7b4f83104f176a6a5f0bef16577617fc2c05395468cc29a02759ca80a639404c","abstract_canon_sha256":"e78a234df3fff9e89b0b9d316463df8d95269fa8f5e7cd871350a2b831fcbbe0"},"schema_version":"1.0"},"canonical_sha256":"5ec59786a12d1d3e0d76056dcfc38546083fc2b0dafdf993c5c84dfe5611a738","source":{"kind":"arxiv","id":"2606.04807","version":1},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2606.04807","created_at":"2026-06-04T01:09:30Z"},{"alias_kind":"arxiv_version","alias_value":"2606.04807v1","created_at":"2026-06-04T01:09:30Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2606.04807","created_at":"2026-06-04T01:09:30Z"},{"alias_kind":"pith_short_12","alias_value":"L3CZPBVBFUOT","created_at":"2026-06-04T01:09:30Z"},{"alias_kind":"pith_short_16","alias_value":"L3CZPBVBFUOT4DLW","created_at":"2026-06-04T01:09:30Z"},{"alias_kind":"pith_short_8","alias_value":"L3CZPBVB","created_at":"2026-06-04T01:09:30Z"}],"events":[{"event_type":"record_created","subject_pith_number":"pith:2026:L3CZPBVBFUOT4DLWAVW47Q4FIY","target":"record","payload":{"canonical_record":{"source":{"id":"2606.04807","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.AI","submitted_at":"2026-06-03T12:31:42Z","cross_cats_sorted":["cs.CL","cs.CY","cs.LG"],"title_canon_sha256":"7b4f83104f176a6a5f0bef16577617fc2c05395468cc29a02759ca80a639404c","abstract_canon_sha256":"e78a234df3fff9e89b0b9d316463df8d95269fa8f5e7cd871350a2b831fcbbe0"},"schema_version":"1.0"},"canonical_sha256":"5ec59786a12d1d3e0d76056dcfc38546083fc2b0dafdf993c5c84dfe5611a738","receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-06-04T01:09:30.780442Z","signature_b64":"WngCZcHg3veCJd5nLN8yXDCercUtmXmLFFmCGj6K4wp+X1wYaJk2f3HaUFJnaR0g1WMr/zMHhaKuO+ZAhDL3Bg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"5ec59786a12d1d3e0d76056dcfc38546083fc2b0dafdf993c5c84dfe5611a738","last_reissued_at":"2026-06-04T01:09:30.779712Z","signature_status":"signed_v1","first_computed_at":"2026-06-04T01:09:30.779712Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"source_kind":"arxiv","source_id":"2606.04807","source_version":1,"attestation_state":"computed"},"signer":{"signer_id":"pith.science","signer_type":"pith_registry","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"created_at":"2026-06-04T01:09:30Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"e8H5BAbW9OFbAPt4o7qpmSLVZs0w8SA8zAUhDH6X2KH4rNIkVu/HOg7NKJhUSP/e5LRRIsV6RgPXeA58jZKAAw==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-07-03T16:47:40.306378Z"},"content_sha256":"8a6829007e37498779c985d1fbdaab119b7137c15036977ba54326c8df8236d1","schema_version":"1.0","event_id":"sha256:8a6829007e37498779c985d1fbdaab119b7137c15036977ba54326c8df8236d1"},{"event_type":"graph_snapshot","subject_pith_number":"pith:2026:L3CZPBVBFUOT4DLWAVW47Q4FIY","target":"graph","payload":{"graph_snapshot":{"paper":{"title":"BiasGRPO: Stabilizing Bias Mitigation in High-Variance Reward Landscapes via Group-Relative Policy Optimization","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.CL","cs.CY","cs.LG"],"primary_cat":"cs.AI","authors_text":"ChengXiang Zhai, Ke Yang, Saket Reddy","submitted_at":"2026-06-03T12:31:42Z","abstract_excerpt":"Mitigating social bias in Large Language Models (LLMs) presents a distinct alignment challenge: unlike verifiable tasks, bias lacks a single ground truth, creating a high-variance, subjective reward landscape. Previous preference-based fine-tuning methods have major trade-offs: Direct Preference Optimization (DPO) is limited by the lack of exploration inherent in offline training, while Proximal Policy Optimization (PPO) can lead to training instability due to potentially unreliable critic estimates. In this paper, we propose BiasGRPO, a framework using Group Relative Policy Optimization (GRPO"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2606.04807","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/2606.04807/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":"2026-06-04T01:09:30Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"9zghXLOhmxDaVVPL1Bn7CLBJgZRwx/wcV/S7rxFvXVa0+pYUQXUDu4AXtquSIAiVLf9d1tH897DmZMLDRPZyCA==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-07-03T16:47:40.306765Z"},"content_sha256":"878ef3163c43fb4da4d3c3738f375c3e545dae1c1f573d7cf987534d23fb847b","schema_version":"1.0","event_id":"sha256:878ef3163c43fb4da4d3c3738f375c3e545dae1c1f573d7cf987534d23fb847b"}],"timestamp_proofs":[],"mirror_hints":[{"mirror_type":"https","name":"Pith Resolver","base_url":"https://pith.science","bundle_url":"https://pith.science/pith/L3CZPBVBFUOT4DLWAVW47Q4FIY/bundle.json","state_url":"https://pith.science/pith/L3CZPBVBFUOT4DLWAVW47Q4FIY/state.json","well_known_bundle_url":"https://pith.science/.well-known/pith/L3CZPBVBFUOT4DLWAVW47Q4FIY/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":"2026-07-03T16:47:40Z","links":{"resolver":"https://pith.science/pith/L3CZPBVBFUOT4DLWAVW47Q4FIY","bundle":"https://pith.science/pith/L3CZPBVBFUOT4DLWAVW47Q4FIY/bundle.json","state":"https://pith.science/pith/L3CZPBVBFUOT4DLWAVW47Q4FIY/state.json","well_known_bundle":"https://pith.science/.well-known/pith/L3CZPBVBFUOT4DLWAVW47Q4FIY/bundle.json"},"state":{"state_type":"pith_open_graph_state","state_version":"1.0","pith_number":"pith:2026:L3CZPBVBFUOT4DLWAVW47Q4FIY","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":"e78a234df3fff9e89b0b9d316463df8d95269fa8f5e7cd871350a2b831fcbbe0","cross_cats_sorted":["cs.CL","cs.CY","cs.LG"],"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.AI","submitted_at":"2026-06-03T12:31:42Z","title_canon_sha256":"7b4f83104f176a6a5f0bef16577617fc2c05395468cc29a02759ca80a639404c"},"schema_version":"1.0","source":{"id":"2606.04807","kind":"arxiv","version":1}},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2606.04807","created_at":"2026-06-04T01:09:30Z"},{"alias_kind":"arxiv_version","alias_value":"2606.04807v1","created_at":"2026-06-04T01:09:30Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2606.04807","created_at":"2026-06-04T01:09:30Z"},{"alias_kind":"pith_short_12","alias_value":"L3CZPBVBFUOT","created_at":"2026-06-04T01:09:30Z"},{"alias_kind":"pith_short_16","alias_value":"L3CZPBVBFUOT4DLW","created_at":"2026-06-04T01:09:30Z"},{"alias_kind":"pith_short_8","alias_value":"L3CZPBVB","created_at":"2026-06-04T01:09:30Z"}],"graph_snapshots":[{"event_id":"sha256:878ef3163c43fb4da4d3c3738f375c3e545dae1c1f573d7cf987534d23fb847b","target":"graph","created_at":"2026-06-04T01:09:30Z","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/2606.04807/integrity.json","findings":[],"snapshot_sha256":"c28c3603d3b5d939e8dc4c7e95fa8dfce3d595e45f758748cecf8e644a296938","summary":{"advisory":0,"by_detector":{},"critical":0,"informational":0}},"paper":{"abstract_excerpt":"Mitigating social bias in Large Language Models (LLMs) presents a distinct alignment challenge: unlike verifiable tasks, bias lacks a single ground truth, creating a high-variance, subjective reward landscape. Previous preference-based fine-tuning methods have major trade-offs: Direct Preference Optimization (DPO) is limited by the lack of exploration inherent in offline training, while Proximal Policy Optimization (PPO) can lead to training instability due to potentially unreliable critic estimates. In this paper, we propose BiasGRPO, a framework using Group Relative Policy Optimization (GRPO","authors_text":"ChengXiang Zhai, Ke Yang, Saket Reddy","cross_cats":["cs.CL","cs.CY","cs.LG"],"headline":"","license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.AI","submitted_at":"2026-06-03T12:31:42Z","title":"BiasGRPO: Stabilizing Bias Mitigation in High-Variance Reward Landscapes via Group-Relative Policy Optimization"},"references":{"count":0,"internal_anchors":0,"resolved_work":0,"sample":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2606.04807","kind":"arxiv","version":1},"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":"sha256:8a6829007e37498779c985d1fbdaab119b7137c15036977ba54326c8df8236d1","target":"record","created_at":"2026-06-04T01:09:30Z","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":"e78a234df3fff9e89b0b9d316463df8d95269fa8f5e7cd871350a2b831fcbbe0","cross_cats_sorted":["cs.CL","cs.CY","cs.LG"],"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.AI","submitted_at":"2026-06-03T12:31:42Z","title_canon_sha256":"7b4f83104f176a6a5f0bef16577617fc2c05395468cc29a02759ca80a639404c"},"schema_version":"1.0","source":{"id":"2606.04807","kind":"arxiv","version":1}},"canonical_sha256":"5ec59786a12d1d3e0d76056dcfc38546083fc2b0dafdf993c5c84dfe5611a738","receipt":{"algorithm":"ed25519","builder_version":"pith-number-builder-2026-05-17-v1","canonical_sha256":"5ec59786a12d1d3e0d76056dcfc38546083fc2b0dafdf993c5c84dfe5611a738","first_computed_at":"2026-06-04T01:09:30.779712Z","key_id":"pith-v1-2026-05","kind":"pith_receipt","last_reissued_at":"2026-06-04T01:09:30.779712Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","receipt_version":"0.3","signature_b64":"WngCZcHg3veCJd5nLN8yXDCercUtmXmLFFmCGj6K4wp+X1wYaJk2f3HaUFJnaR0g1WMr/zMHhaKuO+ZAhDL3Bg==","signature_status":"signed_v1","signed_at":"2026-06-04T01:09:30.780442Z","signed_message":"canonical_sha256_bytes"},"source_id":"2606.04807","source_kind":"arxiv","source_version":1}}},"equivocations":[],"invalid_events":[],"applied_event_ids":["sha256:8a6829007e37498779c985d1fbdaab119b7137c15036977ba54326c8df8236d1","sha256:878ef3163c43fb4da4d3c3738f375c3e545dae1c1f573d7cf987534d23fb847b"],"state_sha256":"31e2d082e43f8ea3fb10b71be6c664193862b9f806132d2750b450b82d546eea"},"bundle_signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"/O6d1A/7+/9iPlhZO2qaXDdXiYxHEL/4Gbii4t4ceH4l603ZZAPPYuQ5C0PA7QwRXZHK9qPOrUxTtbyGOmN3Dg==","signed_message":"bundle_sha256_bytes","signed_at":"2026-07-03T16:47:40.308949Z","bundle_sha256":"332f910e07665176b063c7aa3ccd4a4ae4676f012b8342f50fa39efd628d25cb"}}