{"bundle_type":"pith_open_graph_bundle","bundle_version":"1.0","pith_number":"pith:2025:J33L7ABFSXLPLYSICON4C2XC7B","short_pith_number":"pith:J33L7ABF","canonical_record":{"source":{"id":"2506.10922","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2025-06-12T17:34:38Z","cross_cats_sorted":["cs.AI","cs.CL"],"title_canon_sha256":"93d07e77f8681845c80b46fd87e32d27ac2081d167fee39fef46082a8264efb9","abstract_canon_sha256":"7b702477d588ef08385e9c4e26054025de9e91c66708d87f06a2b3f245c34348"},"schema_version":"1.0"},"canonical_sha256":"4ef6bf802595d6f5e248139bc16ae2f8611f1a8e1900e9c2d4308ea4fcbfe547","source":{"kind":"arxiv","id":"2506.10922","version":1},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2506.10922","created_at":"2026-07-05T11:20:40Z"},{"alias_kind":"arxiv_version","alias_value":"2506.10922v1","created_at":"2026-07-05T11:20:40Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2506.10922","created_at":"2026-07-05T11:20:40Z"},{"alias_kind":"pith_short_12","alias_value":"J33L7ABFSXLP","created_at":"2026-07-05T11:20:40Z"},{"alias_kind":"pith_short_16","alias_value":"J33L7ABFSXLPLYSI","created_at":"2026-07-05T11:20:40Z"},{"alias_kind":"pith_short_8","alias_value":"J33L7ABF","created_at":"2026-07-05T11:20:40Z"}],"events":[{"event_type":"record_created","subject_pith_number":"pith:2025:J33L7ABFSXLPLYSICON4C2XC7B","target":"record","payload":{"canonical_record":{"source":{"id":"2506.10922","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2025-06-12T17:34:38Z","cross_cats_sorted":["cs.AI","cs.CL"],"title_canon_sha256":"93d07e77f8681845c80b46fd87e32d27ac2081d167fee39fef46082a8264efb9","abstract_canon_sha256":"7b702477d588ef08385e9c4e26054025de9e91c66708d87f06a2b3f245c34348"},"schema_version":"1.0"},"canonical_sha256":"4ef6bf802595d6f5e248139bc16ae2f8611f1a8e1900e9c2d4308ea4fcbfe547","receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:20:40.400348Z","signature_b64":"ay1L2+sCGiz6P3HR7rpUpaPsWcosMkAutx85e7KZm5AM+P/PrNhWSu/DT3OK6fM74hgU+SYsD6FP7+RYG0qDAg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"4ef6bf802595d6f5e248139bc16ae2f8611f1a8e1900e9c2d4308ea4fcbfe547","last_reissued_at":"2026-07-05T11:20:40.399833Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:20:40.399833Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"source_kind":"arxiv","source_id":"2506.10922","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":"2026-07-05T11:20:40Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"Q2Iiqkn1A0cXoQ9+fMxJJhTUQqiJt5LtRWZA/V6LpbdbCbAofNhfIIqKyobXOUUOpcG6NLGJCxu46W9G68ALCw==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-08-07T22:34:02.027334Z"},"content_sha256":"f87fbbab4e2f39adf3a655da102ae5fc260436e0c498fa0ebc6059458f05a2cf","schema_version":"1.0","event_id":"sha256:f87fbbab4e2f39adf3a655da102ae5fc260436e0c498fa0ebc6059458f05a2cf"},{"event_type":"graph_snapshot","subject_pith_number":"pith:2025:J33L7ABFSXLPLYSICON4C2XC7B","target":"graph","payload":{"graph_snapshot":{"paper":{"title":"Robustly Improving LLM Fairness in Realistic Settings via Interpretability","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI","cs.CL"],"primary_cat":"cs.LG","authors_text":"Adam Karvonen, Samuel Marks","submitted_at":"2025-06-12T17:34:38Z","abstract_excerpt":"Large language models (LLMs) are increasingly deployed in high-stakes hiring applications, making decisions that directly impact people's careers and livelihoods. While prior studies suggest simple anti-bias prompts can eliminate demographic biases in controlled evaluations, we find these mitigations fail when realistic contextual details are introduced. We address these failures through internal bias mitigation: by identifying and neutralizing sensitive attribute directions within model activations, we achieve robust bias reduction across all tested scenarios. Across leading commercial (GPT-4"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2506.10922","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/2506.10922/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":"2026-07-05T11:20:40Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"gc7NUX48z1sRyd2eYpxSKmaCah6HxpZ66VXeUp2TnhHFMneqTQMc45jno8I4bDqHSguNBTCtgGBM0c4cmVgQBA==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-08-07T22:34:02.027851Z"},"content_sha256":"14e92de1cf398680284762c7abad02a665ef1c7632dfb6c716bb1c8046dbac29","schema_version":"1.0","event_id":"sha256:14e92de1cf398680284762c7abad02a665ef1c7632dfb6c716bb1c8046dbac29"}],"timestamp_proofs":[],"mirror_hints":[{"mirror_type":"https","name":"Pith Resolver","base_url":"https://pith.science","bundle_url":"https://pith.science/pith/J33L7ABFSXLPLYSICON4C2XC7B/bundle.json","state_url":"https://pith.science/pith/J33L7ABFSXLPLYSICON4C2XC7B/state.json","well_known_bundle_url":"https://pith.science/.well-known/pith/J33L7ABFSXLPLYSICON4C2XC7B/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":"2026-08-07T22:34:02Z","links":{"resolver":"https://pith.science/pith/J33L7ABFSXLPLYSICON4C2XC7B","bundle":"https://pith.science/pith/J33L7ABFSXLPLYSICON4C2XC7B/bundle.json","state":"https://pith.science/pith/J33L7ABFSXLPLYSICON4C2XC7B/state.json","well_known_bundle":"https://pith.science/.well-known/pith/J33L7ABFSXLPLYSICON4C2XC7B/bundle.json"},"state":{"state_type":"pith_open_graph_state","state_version":"1.0","pith_number":"pith:2025:J33L7ABFSXLPLYSICON4C2XC7B","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":"7b702477d588ef08385e9c4e26054025de9e91c66708d87f06a2b3f245c34348","cross_cats_sorted":["cs.AI","cs.CL"],"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2025-06-12T17:34:38Z","title_canon_sha256":"93d07e77f8681845c80b46fd87e32d27ac2081d167fee39fef46082a8264efb9"},"schema_version":"1.0","source":{"id":"2506.10922","kind":"arxiv","version":1}},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2506.10922","created_at":"2026-07-05T11:20:40Z"},{"alias_kind":"arxiv_version","alias_value":"2506.10922v1","created_at":"2026-07-05T11:20:40Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2506.10922","created_at":"2026-07-05T11:20:40Z"},{"alias_kind":"pith_short_12","alias_value":"J33L7ABFSXLP","created_at":"2026-07-05T11:20:40Z"},{"alias_kind":"pith_short_16","alias_value":"J33L7ABFSXLPLYSI","created_at":"2026-07-05T11:20:40Z"},{"alias_kind":"pith_short_8","alias_value":"J33L7ABF","created_at":"2026-07-05T11:20:40Z"}],"graph_snapshots":[{"event_id":"sha256:14e92de1cf398680284762c7abad02a665ef1c7632dfb6c716bb1c8046dbac29","target":"graph","created_at":"2026-07-05T11:20:40Z","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/2506.10922/integrity.json","findings":[],"snapshot_sha256":"c28c3603d3b5d939e8dc4c7e95fa8dfce3d595e45f758748cecf8e644a296938","summary":{"advisory":0,"by_detector":{},"critical":0,"informational":0}},"paper":{"abstract_excerpt":"Large language models (LLMs) are increasingly deployed in high-stakes hiring applications, making decisions that directly impact people's careers and livelihoods. While prior studies suggest simple anti-bias prompts can eliminate demographic biases in controlled evaluations, we find these mitigations fail when realistic contextual details are introduced. We address these failures through internal bias mitigation: by identifying and neutralizing sensitive attribute directions within model activations, we achieve robust bias reduction across all tested scenarios. Across leading commercial (GPT-4","authors_text":"Adam Karvonen, Samuel Marks","cross_cats":["cs.AI","cs.CL"],"headline":"","license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2025-06-12T17:34:38Z","title":"Robustly Improving LLM Fairness in Realistic Settings via Interpretability"},"references":{"count":0,"internal_anchors":0,"resolved_work":0,"sample":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2506.10922","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:f87fbbab4e2f39adf3a655da102ae5fc260436e0c498fa0ebc6059458f05a2cf","target":"record","created_at":"2026-07-05T11:20:40Z","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":"7b702477d588ef08385e9c4e26054025de9e91c66708d87f06a2b3f245c34348","cross_cats_sorted":["cs.AI","cs.CL"],"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2025-06-12T17:34:38Z","title_canon_sha256":"93d07e77f8681845c80b46fd87e32d27ac2081d167fee39fef46082a8264efb9"},"schema_version":"1.0","source":{"id":"2506.10922","kind":"arxiv","version":1}},"canonical_sha256":"4ef6bf802595d6f5e248139bc16ae2f8611f1a8e1900e9c2d4308ea4fcbfe547","receipt":{"algorithm":"ed25519","builder_version":"pith-number-builder-2026-05-17-v1","canonical_sha256":"4ef6bf802595d6f5e248139bc16ae2f8611f1a8e1900e9c2d4308ea4fcbfe547","first_computed_at":"2026-07-05T11:20:40.399833Z","key_id":"pith-v1-2026-05","kind":"pith_receipt","last_reissued_at":"2026-07-05T11:20:40.399833Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","receipt_version":"0.3","signature_b64":"ay1L2+sCGiz6P3HR7rpUpaPsWcosMkAutx85e7KZm5AM+P/PrNhWSu/DT3OK6fM74hgU+SYsD6FP7+RYG0qDAg==","signature_status":"signed_v1","signed_at":"2026-07-05T11:20:40.400348Z","signed_message":"canonical_sha256_bytes"},"source_id":"2506.10922","source_kind":"arxiv","source_version":1}}},"equivocations":[],"invalid_events":[],"applied_event_ids":["sha256:f87fbbab4e2f39adf3a655da102ae5fc260436e0c498fa0ebc6059458f05a2cf","sha256:14e92de1cf398680284762c7abad02a665ef1c7632dfb6c716bb1c8046dbac29"],"state_sha256":"b8abd9f05bf610bde1a1326bf88eff7b4a6552f4ab6b7d3f457a5561cdb346d6"},"bundle_signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"WD+SAn9kInprk3NlVGtDmuRTh3oDFnIejZotPjtxUGnUjfQVUKvt3DoLldxM8vtndFd9nRYWN/wWvRH1HOerAQ==","signed_message":"bundle_sha256_bytes","signed_at":"2026-08-07T22:34:02.033014Z","bundle_sha256":"0f6a94fd212f13208a0a1a4807936f137a948270744d54b3328085eb54b403e0"}}