{"bundle_type":"pith_open_graph_bundle","bundle_version":"1.0","pith_number":"pith:2024:N3ZUERXJ7MWIABTNI3WX66I5NH","short_pith_number":"pith:N3ZUERXJ","canonical_record":{"source":{"id":"2409.11704","kind":"arxiv","version":2},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CL","submitted_at":"2024-09-18T05:13:18Z","cross_cats_sorted":["cs.LG"],"title_canon_sha256":"1c4d3cfd1410de5d4f3845efcf49d37baa366636b5a5b37576dc02a1ab6c50cb","abstract_canon_sha256":"8bbee54fb7e3caf780d847a23b97f04cd07a428da8afcd751b7796a7cacfc803"},"schema_version":"1.0"},"canonical_sha256":"6ef34246e9fb2c80066d46ed7f791d69ed739d534c053577d033c950c263bc73","source":{"kind":"arxiv","id":"2409.11704","version":2},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2409.11704","created_at":"2026-07-05T11:08:00Z"},{"alias_kind":"arxiv_version","alias_value":"2409.11704v2","created_at":"2026-07-05T11:08:00Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2409.11704","created_at":"2026-07-05T11:08:00Z"},{"alias_kind":"pith_short_12","alias_value":"N3ZUERXJ7MWI","created_at":"2026-07-05T11:08:00Z"},{"alias_kind":"pith_short_16","alias_value":"N3ZUERXJ7MWIABTN","created_at":"2026-07-05T11:08:00Z"},{"alias_kind":"pith_short_8","alias_value":"N3ZUERXJ","created_at":"2026-07-05T11:08:00Z"}],"events":[{"event_type":"record_created","subject_pith_number":"pith:2024:N3ZUERXJ7MWIABTNI3WX66I5NH","target":"record","payload":{"canonical_record":{"source":{"id":"2409.11704","kind":"arxiv","version":2},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CL","submitted_at":"2024-09-18T05:13:18Z","cross_cats_sorted":["cs.LG"],"title_canon_sha256":"1c4d3cfd1410de5d4f3845efcf49d37baa366636b5a5b37576dc02a1ab6c50cb","abstract_canon_sha256":"8bbee54fb7e3caf780d847a23b97f04cd07a428da8afcd751b7796a7cacfc803"},"schema_version":"1.0"},"canonical_sha256":"6ef34246e9fb2c80066d46ed7f791d69ed739d534c053577d033c950c263bc73","receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:08:00.786168Z","signature_b64":"iyHCYxT2JozWCvlqMdedDHtQD02gFvw7jlRe5rHgUeQrD1bzT0n7iAytz0SOULJhX8EoMdDhohblNMqzyE6iBw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"6ef34246e9fb2c80066d46ed7f791d69ed739d534c053577d033c950c263bc73","last_reissued_at":"2026-07-05T11:08:00.785668Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:08:00.785668Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"source_kind":"arxiv","source_id":"2409.11704","source_version":2,"attestation_state":"computed"},"signer":{"signer_id":"pith.science","signer_type":"pith_registry","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"created_at":"2026-07-05T11:08:00Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"ztrpc+5BtgsIqazZ5frWOOTIf3aoTk5ykjrqFsHAGK5tqY389Lol4+ngMZPXVAzFJbWJRh2eFVPLNNC1db4tBA==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-08-18T10:32:29.050604Z"},"content_sha256":"fa2eb6e85cf11f87e96f9f4a925e5401441d877382cbaa79b3f2bbb8306ee91e","schema_version":"1.0","event_id":"sha256:fa2eb6e85cf11f87e96f9f4a925e5401441d877382cbaa79b3f2bbb8306ee91e"},{"event_type":"graph_snapshot","subject_pith_number":"pith:2024:N3ZUERXJ7MWIABTNI3WX66I5NH","target":"graph","payload":{"graph_snapshot":{"paper":{"title":"From Lists to Emojis: How Format Bias Affects Model Alignment","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.LG"],"primary_cat":"cs.CL","authors_text":"Heng Huang, Lichang Chen, Tianyi Zhou, Tong Zhang, Wei Xiong, Xuanchang Zhang","submitted_at":"2024-09-18T05:13:18Z","abstract_excerpt":"In this paper, we study format biases in reinforcement learning from human feedback (RLHF). We observe that many widely-used preference models, including human evaluators, GPT-4, and top-ranking models on the RewardBench benchmark, exhibit strong biases towards specific format patterns, such as lists, links, bold text, and emojis. Furthermore, large language models (LLMs) can exploit these biases to achieve higher rankings on popular benchmarks like AlpacaEval and LMSYS Chatbot Arena. One notable example of this is verbosity bias, where current preference models favor longer responses that app"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2409.11704","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/2409.11704/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":"2026-07-05T11:08:00Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"PaQ8/Y8dJM2KmiRXgNHf34wst7ZgIRfeu2uM5PnRXmSPwxNvWDoALly9Jh0wRODoVOFMxW0UraZbx0v9ehXLBQ==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-08-18T10:32:29.051234Z"},"content_sha256":"0b0443d31c755a2ac1154a38ec8ae8b6a551c2fcbc3d0acf09edfd44bafbd78a","schema_version":"1.0","event_id":"sha256:0b0443d31c755a2ac1154a38ec8ae8b6a551c2fcbc3d0acf09edfd44bafbd78a"}],"timestamp_proofs":[],"mirror_hints":[{"mirror_type":"https","name":"Pith Resolver","base_url":"https://pith.science","bundle_url":"https://pith.science/pith/N3ZUERXJ7MWIABTNI3WX66I5NH/bundle.json","state_url":"https://pith.science/pith/N3ZUERXJ7MWIABTNI3WX66I5NH/state.json","well_known_bundle_url":"https://pith.science/.well-known/pith/N3ZUERXJ7MWIABTNI3WX66I5NH/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":"2026-08-18T10:32:29Z","links":{"resolver":"https://pith.science/pith/N3ZUERXJ7MWIABTNI3WX66I5NH","bundle":"https://pith.science/pith/N3ZUERXJ7MWIABTNI3WX66I5NH/bundle.json","state":"https://pith.science/pith/N3ZUERXJ7MWIABTNI3WX66I5NH/state.json","well_known_bundle":"https://pith.science/.well-known/pith/N3ZUERXJ7MWIABTNI3WX66I5NH/bundle.json"},"state":{"state_type":"pith_open_graph_state","state_version":"1.0","pith_number":"pith:2024:N3ZUERXJ7MWIABTNI3WX66I5NH","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":"8bbee54fb7e3caf780d847a23b97f04cd07a428da8afcd751b7796a7cacfc803","cross_cats_sorted":["cs.LG"],"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CL","submitted_at":"2024-09-18T05:13:18Z","title_canon_sha256":"1c4d3cfd1410de5d4f3845efcf49d37baa366636b5a5b37576dc02a1ab6c50cb"},"schema_version":"1.0","source":{"id":"2409.11704","kind":"arxiv","version":2}},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2409.11704","created_at":"2026-07-05T11:08:00Z"},{"alias_kind":"arxiv_version","alias_value":"2409.11704v2","created_at":"2026-07-05T11:08:00Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2409.11704","created_at":"2026-07-05T11:08:00Z"},{"alias_kind":"pith_short_12","alias_value":"N3ZUERXJ7MWI","created_at":"2026-07-05T11:08:00Z"},{"alias_kind":"pith_short_16","alias_value":"N3ZUERXJ7MWIABTN","created_at":"2026-07-05T11:08:00Z"},{"alias_kind":"pith_short_8","alias_value":"N3ZUERXJ","created_at":"2026-07-05T11:08:00Z"}],"graph_snapshots":[{"event_id":"sha256:0b0443d31c755a2ac1154a38ec8ae8b6a551c2fcbc3d0acf09edfd44bafbd78a","target":"graph","created_at":"2026-07-05T11:08:00Z","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/2409.11704/integrity.json","findings":[],"snapshot_sha256":"c28c3603d3b5d939e8dc4c7e95fa8dfce3d595e45f758748cecf8e644a296938","summary":{"advisory":0,"by_detector":{},"critical":0,"informational":0}},"paper":{"abstract_excerpt":"In this paper, we study format biases in reinforcement learning from human feedback (RLHF). We observe that many widely-used preference models, including human evaluators, GPT-4, and top-ranking models on the RewardBench benchmark, exhibit strong biases towards specific format patterns, such as lists, links, bold text, and emojis. Furthermore, large language models (LLMs) can exploit these biases to achieve higher rankings on popular benchmarks like AlpacaEval and LMSYS Chatbot Arena. One notable example of this is verbosity bias, where current preference models favor longer responses that app","authors_text":"Heng Huang, Lichang Chen, Tianyi Zhou, Tong Zhang, Wei Xiong, Xuanchang Zhang","cross_cats":["cs.LG"],"headline":"","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CL","submitted_at":"2024-09-18T05:13:18Z","title":"From Lists to Emojis: How Format Bias Affects Model Alignment"},"references":{"count":0,"internal_anchors":0,"resolved_work":0,"sample":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2409.11704","kind":"arxiv","version":2},"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:fa2eb6e85cf11f87e96f9f4a925e5401441d877382cbaa79b3f2bbb8306ee91e","target":"record","created_at":"2026-07-05T11:08:00Z","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":"8bbee54fb7e3caf780d847a23b97f04cd07a428da8afcd751b7796a7cacfc803","cross_cats_sorted":["cs.LG"],"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CL","submitted_at":"2024-09-18T05:13:18Z","title_canon_sha256":"1c4d3cfd1410de5d4f3845efcf49d37baa366636b5a5b37576dc02a1ab6c50cb"},"schema_version":"1.0","source":{"id":"2409.11704","kind":"arxiv","version":2}},"canonical_sha256":"6ef34246e9fb2c80066d46ed7f791d69ed739d534c053577d033c950c263bc73","receipt":{"algorithm":"ed25519","builder_version":"pith-number-builder-2026-05-17-v1","canonical_sha256":"6ef34246e9fb2c80066d46ed7f791d69ed739d534c053577d033c950c263bc73","first_computed_at":"2026-07-05T11:08:00.785668Z","key_id":"pith-v1-2026-05","kind":"pith_receipt","last_reissued_at":"2026-07-05T11:08:00.785668Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","receipt_version":"0.3","signature_b64":"iyHCYxT2JozWCvlqMdedDHtQD02gFvw7jlRe5rHgUeQrD1bzT0n7iAytz0SOULJhX8EoMdDhohblNMqzyE6iBw==","signature_status":"signed_v1","signed_at":"2026-07-05T11:08:00.786168Z","signed_message":"canonical_sha256_bytes"},"source_id":"2409.11704","source_kind":"arxiv","source_version":2}}},"equivocations":[],"invalid_events":[],"applied_event_ids":["sha256:fa2eb6e85cf11f87e96f9f4a925e5401441d877382cbaa79b3f2bbb8306ee91e","sha256:0b0443d31c755a2ac1154a38ec8ae8b6a551c2fcbc3d0acf09edfd44bafbd78a"],"state_sha256":"80384f317ca95f6c916f3d163257e0757e75f841fa6551fa3e54d79f40638114"},"bundle_signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"O1JmEkYiOBU9ppFgnkZoB+JF0gRmZ1mrIrdHdlFrYq2dwJByItKZZ7UYh4/CC4v2WO1vDZGoozKk7aJy/IrxAQ==","signed_message":"bundle_sha256_bytes","signed_at":"2026-08-18T10:32:29.056872Z","bundle_sha256":"bec9f9eb44607b4b520f99b7649f98cb04ca5af1bd05855b2ca8e85d85f535fd"}}