{"bundle_type":"pith_open_graph_bundle","bundle_version":"1.0","pith_number":"pith:2024:UBHXHTNMYQTOMWZZCU2HE34ZG7","short_pith_number":"pith:UBHXHTNM","canonical_record":{"source":{"id":"2407.04549","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CL","submitted_at":"2024-07-05T14:34:50Z","cross_cats_sorted":["cs.AI"],"title_canon_sha256":"39de3a4dbec115a35277f172e74df78cfb6b2ce20a36afdfe6827c2aa1a0fd19","abstract_canon_sha256":"01eb9de96a300345f94b75770590ec33d10d1bc20ab0ce36cc7c9211bb21ab1f"},"schema_version":"1.0"},"canonical_sha256":"a04f73cdacc426e65b391534726f9937c87104f307225c6834fad7d28a780df5","source":{"kind":"arxiv","id":"2407.04549","version":1},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2407.04549","created_at":"2026-07-05T08:40:35Z"},{"alias_kind":"arxiv_version","alias_value":"2407.04549v1","created_at":"2026-07-05T08:40:35Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2407.04549","created_at":"2026-07-05T08:40:35Z"},{"alias_kind":"pith_short_12","alias_value":"UBHXHTNMYQTO","created_at":"2026-07-05T08:40:35Z"},{"alias_kind":"pith_short_16","alias_value":"UBHXHTNMYQTOMWZZ","created_at":"2026-07-05T08:40:35Z"},{"alias_kind":"pith_short_8","alias_value":"UBHXHTNM","created_at":"2026-07-05T08:40:35Z"}],"events":[{"event_type":"record_created","subject_pith_number":"pith:2024:UBHXHTNMYQTOMWZZCU2HE34ZG7","target":"record","payload":{"canonical_record":{"source":{"id":"2407.04549","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CL","submitted_at":"2024-07-05T14:34:50Z","cross_cats_sorted":["cs.AI"],"title_canon_sha256":"39de3a4dbec115a35277f172e74df78cfb6b2ce20a36afdfe6827c2aa1a0fd19","abstract_canon_sha256":"01eb9de96a300345f94b75770590ec33d10d1bc20ab0ce36cc7c9211bb21ab1f"},"schema_version":"1.0"},"canonical_sha256":"a04f73cdacc426e65b391534726f9937c87104f307225c6834fad7d28a780df5","receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T08:40:35.367725Z","signature_b64":"vnfT5F7vRg1V1TNXimBCYmgntVxpT3aWp336f6twHPQ1kpO3EbUgJxa2HwCMRftQKZfQk5BF6GHKoL4Wa1r/Aw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"a04f73cdacc426e65b391534726f9937c87104f307225c6834fad7d28a780df5","last_reissued_at":"2026-07-05T08:40:35.367248Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T08:40:35.367248Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"source_kind":"arxiv","source_id":"2407.04549","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":"2026-07-05T08:40:35Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"4E5ZeQfZCEq73TYUVk3+NqLx3pdWvFcXc6BTA4Sw8CZUAx8x4ICJhG9HVgj0zy4gS306ErBNBrFYm27ED5YeAw==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-08-22T13:19:40.681086Z"},"content_sha256":"1a69afc8971d2c0d6bc6664e66dbea55b6278da5e649f9f9ae282f08c952df92","schema_version":"1.0","event_id":"sha256:1a69afc8971d2c0d6bc6664e66dbea55b6278da5e649f9f9ae282f08c952df92"},{"event_type":"graph_snapshot","subject_pith_number":"pith:2024:UBHXHTNMYQTOMWZZCU2HE34ZG7","target":"graph","payload":{"graph_snapshot":{"paper":{"title":"Spontaneous Reward Hacking in Iterative Self-Refinement","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.CL","authors_text":"He He, Jane Pan, Samuel R. Bowman, Shi Feng","submitted_at":"2024-07-05T14:34:50Z","abstract_excerpt":"Language models are capable of iteratively improving their outputs based on natural language feedback, thus enabling in-context optimization of user preference. In place of human users, a second language model can be used as an evaluator, providing feedback along with numerical ratings which the generator attempts to optimize. However, because the evaluator is an imperfect proxy of user preference, this optimization can lead to reward hacking, where the evaluator's ratings improve while the generation quality remains stagnant or even decreases as judged by actual user preference. The concern o"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2407.04549","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/2407.04549/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":"2026-07-05T08:40:35Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"vHjDcmMThr2e8r7Cre0b6nPHRToS3KnUQk4857NSIPeMN9IHoMZfMUdyEQuvoc5KvBI/UQqefg5hQyNeBrkwDw==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-08-22T13:19:40.681587Z"},"content_sha256":"ba09b1c3538b1ced0f3a8c323ce011f14683063d9bb04c6043ea5ea6f213f27e","schema_version":"1.0","event_id":"sha256:ba09b1c3538b1ced0f3a8c323ce011f14683063d9bb04c6043ea5ea6f213f27e"}],"timestamp_proofs":[],"mirror_hints":[{"mirror_type":"https","name":"Pith Resolver","base_url":"https://pith.science","bundle_url":"https://pith.science/pith/UBHXHTNMYQTOMWZZCU2HE34ZG7/bundle.json","state_url":"https://pith.science/pith/UBHXHTNMYQTOMWZZCU2HE34ZG7/state.json","well_known_bundle_url":"https://pith.science/.well-known/pith/UBHXHTNMYQTOMWZZCU2HE34ZG7/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":"2026-08-22T13:19:40Z","links":{"resolver":"https://pith.science/pith/UBHXHTNMYQTOMWZZCU2HE34ZG7","bundle":"https://pith.science/pith/UBHXHTNMYQTOMWZZCU2HE34ZG7/bundle.json","state":"https://pith.science/pith/UBHXHTNMYQTOMWZZCU2HE34ZG7/state.json","well_known_bundle":"https://pith.science/.well-known/pith/UBHXHTNMYQTOMWZZCU2HE34ZG7/bundle.json"},"state":{"state_type":"pith_open_graph_state","state_version":"1.0","pith_number":"pith:2024:UBHXHTNMYQTOMWZZCU2HE34ZG7","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":"01eb9de96a300345f94b75770590ec33d10d1bc20ab0ce36cc7c9211bb21ab1f","cross_cats_sorted":["cs.AI"],"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CL","submitted_at":"2024-07-05T14:34:50Z","title_canon_sha256":"39de3a4dbec115a35277f172e74df78cfb6b2ce20a36afdfe6827c2aa1a0fd19"},"schema_version":"1.0","source":{"id":"2407.04549","kind":"arxiv","version":1}},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2407.04549","created_at":"2026-07-05T08:40:35Z"},{"alias_kind":"arxiv_version","alias_value":"2407.04549v1","created_at":"2026-07-05T08:40:35Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2407.04549","created_at":"2026-07-05T08:40:35Z"},{"alias_kind":"pith_short_12","alias_value":"UBHXHTNMYQTO","created_at":"2026-07-05T08:40:35Z"},{"alias_kind":"pith_short_16","alias_value":"UBHXHTNMYQTOMWZZ","created_at":"2026-07-05T08:40:35Z"},{"alias_kind":"pith_short_8","alias_value":"UBHXHTNM","created_at":"2026-07-05T08:40:35Z"}],"graph_snapshots":[{"event_id":"sha256:ba09b1c3538b1ced0f3a8c323ce011f14683063d9bb04c6043ea5ea6f213f27e","target":"graph","created_at":"2026-07-05T08:40:35Z","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/2407.04549/integrity.json","findings":[],"snapshot_sha256":"c28c3603d3b5d939e8dc4c7e95fa8dfce3d595e45f758748cecf8e644a296938","summary":{"advisory":0,"by_detector":{},"critical":0,"informational":0}},"paper":{"abstract_excerpt":"Language models are capable of iteratively improving their outputs based on natural language feedback, thus enabling in-context optimization of user preference. In place of human users, a second language model can be used as an evaluator, providing feedback along with numerical ratings which the generator attempts to optimize. However, because the evaluator is an imperfect proxy of user preference, this optimization can lead to reward hacking, where the evaluator's ratings improve while the generation quality remains stagnant or even decreases as judged by actual user preference. The concern o","authors_text":"He He, Jane Pan, Samuel R. Bowman, Shi Feng","cross_cats":["cs.AI"],"headline":"","license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CL","submitted_at":"2024-07-05T14:34:50Z","title":"Spontaneous Reward Hacking in Iterative Self-Refinement"},"references":{"count":0,"internal_anchors":0,"resolved_work":0,"sample":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2407.04549","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":"sha256:1a69afc8971d2c0d6bc6664e66dbea55b6278da5e649f9f9ae282f08c952df92","target":"record","created_at":"2026-07-05T08:40:35Z","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":"01eb9de96a300345f94b75770590ec33d10d1bc20ab0ce36cc7c9211bb21ab1f","cross_cats_sorted":["cs.AI"],"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CL","submitted_at":"2024-07-05T14:34:50Z","title_canon_sha256":"39de3a4dbec115a35277f172e74df78cfb6b2ce20a36afdfe6827c2aa1a0fd19"},"schema_version":"1.0","source":{"id":"2407.04549","kind":"arxiv","version":1}},"canonical_sha256":"a04f73cdacc426e65b391534726f9937c87104f307225c6834fad7d28a780df5","receipt":{"algorithm":"ed25519","builder_version":"pith-number-builder-2026-05-17-v1","canonical_sha256":"a04f73cdacc426e65b391534726f9937c87104f307225c6834fad7d28a780df5","first_computed_at":"2026-07-05T08:40:35.367248Z","key_id":"pith-v1-2026-05","kind":"pith_receipt","last_reissued_at":"2026-07-05T08:40:35.367248Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","receipt_version":"0.3","signature_b64":"vnfT5F7vRg1V1TNXimBCYmgntVxpT3aWp336f6twHPQ1kpO3EbUgJxa2HwCMRftQKZfQk5BF6GHKoL4Wa1r/Aw==","signature_status":"signed_v1","signed_at":"2026-07-05T08:40:35.367725Z","signed_message":"canonical_sha256_bytes"},"source_id":"2407.04549","source_kind":"arxiv","source_version":1}}},"equivocations":[],"invalid_events":[],"applied_event_ids":["sha256:1a69afc8971d2c0d6bc6664e66dbea55b6278da5e649f9f9ae282f08c952df92","sha256:ba09b1c3538b1ced0f3a8c323ce011f14683063d9bb04c6043ea5ea6f213f27e"],"state_sha256":"0f2fef9559218f98c4787f63e4a12bdfecb817c1dd2fdae27c2c520bba105074"},"bundle_signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"U2KKLLACrsI68QzttNMg+UsEViQk2uAk7+LAL26yj4cH0H0ZTU9G6W4Y9MlEnjSulZyqopNuAdw5zdY/4Ta/AA==","signed_message":"bundle_sha256_bytes","signed_at":"2026-08-22T13:19:40.685472Z","bundle_sha256":"f5c2fdceece7340fdf11724d8fbf3e0a09727d2d0979e1e11a9031276b426715"}}