{"bundle_type":"pith_open_graph_bundle","bundle_version":"1.0","pith_number":"pith:2023:NLQU5MFX57FPI7VLZGRLRL34TK","short_pith_number":"pith:NLQU5MFX","canonical_record":{"source":{"id":"2312.07551","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by-nc-nd/4.0/","primary_cat":"cs.CL","submitted_at":"2023-12-06T22:53:34Z","cross_cats_sorted":[],"title_canon_sha256":"705b013ebeeeaba997f34a22e95971aaca8510b6af683011b082698f490f4c10","abstract_canon_sha256":"d3912b24c76222ab8f4b10a7988a347bdef379c07163c7ec74366dafee9dac23"},"schema_version":"1.0"},"canonical_sha256":"6ae14eb0b7efcaf47eabc9a2b8af7c9aacca696ac1dd4cae75c4db95a2a12982","source":{"kind":"arxiv","id":"2312.07551","version":1},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2312.07551","created_at":"2026-07-05T07:23:32Z"},{"alias_kind":"arxiv_version","alias_value":"2312.07551v1","created_at":"2026-07-05T07:23:32Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2312.07551","created_at":"2026-07-05T07:23:32Z"},{"alias_kind":"pith_short_12","alias_value":"NLQU5MFX57FP","created_at":"2026-07-05T07:23:32Z"},{"alias_kind":"pith_short_16","alias_value":"NLQU5MFX57FPI7VL","created_at":"2026-07-05T07:23:32Z"},{"alias_kind":"pith_short_8","alias_value":"NLQU5MFX","created_at":"2026-07-05T07:23:32Z"}],"events":[{"event_type":"record_created","subject_pith_number":"pith:2023:NLQU5MFX57FPI7VLZGRLRL34TK","target":"record","payload":{"canonical_record":{"source":{"id":"2312.07551","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by-nc-nd/4.0/","primary_cat":"cs.CL","submitted_at":"2023-12-06T22:53:34Z","cross_cats_sorted":[],"title_canon_sha256":"705b013ebeeeaba997f34a22e95971aaca8510b6af683011b082698f490f4c10","abstract_canon_sha256":"d3912b24c76222ab8f4b10a7988a347bdef379c07163c7ec74366dafee9dac23"},"schema_version":"1.0"},"canonical_sha256":"6ae14eb0b7efcaf47eabc9a2b8af7c9aacca696ac1dd4cae75c4db95a2a12982","receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T07:23:32.576591Z","signature_b64":"uNb9QZjf0FhpcIe06VTXjkH/Q426UfF2C4LuLkb2j2DM9oAXTyZHPhHHXxPcRfZB+MKh3amA2bbliss0aflDBA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"6ae14eb0b7efcaf47eabc9a2b8af7c9aacca696ac1dd4cae75c4db95a2a12982","last_reissued_at":"2026-07-05T07:23:32.576073Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T07:23:32.576073Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"source_kind":"arxiv","source_id":"2312.07551","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":"2026-07-05T07:23:32Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"vMOcPYlOfEh8tODYBFYbevuluFKp9B7+PvDO/mTULkL8Xd/dh/bbJAtA2qNBLU+BfGcncBHFMQRDsggWjAo4BA==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-08-08T11:50:09.129469Z"},"content_sha256":"212f22b2db378dffe703b9dc7a80a8cf9b761b2fffcc6693c1f46c17800991d8","schema_version":"1.0","event_id":"sha256:212f22b2db378dffe703b9dc7a80a8cf9b761b2fffcc6693c1f46c17800991d8"},{"event_type":"graph_snapshot","subject_pith_number":"pith:2023:NLQU5MFX57FPI7VLZGRLRL34TK","target":"graph","payload":{"graph_snapshot":{"paper":{"title":"Language Model Alignment with Elastic Reset","license":"http://creativecommons.org/licenses/by-nc-nd/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Aaron Courville, Florian Strub, Michael Noukhovitch, Samuel Lavoie","submitted_at":"2023-12-06T22:53:34Z","abstract_excerpt":"Finetuning language models with reinforcement learning (RL), e.g. from human feedback (HF), is a prominent method for alignment. But optimizing against a reward model can improve on reward while degrading performance in other areas, a phenomenon known as reward hacking, alignment tax, or language drift. First, we argue that commonly-used test metrics are insufficient and instead measure how different algorithms tradeoff between reward and drift. The standard method modified the reward with a Kullback-Lieber (KL) penalty between the online and initial model. We propose Elastic Reset, a new algo"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2312.07551","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/2312.07551/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":"2026-07-05T07:23:32Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"Fe076dkGcAa+YBz9RPQTm5LNJ+m7t58vq76bLPaUHCBuuSszcDS9eFsBmJl6QxLt+WozOBz1Sz4pJv15tN4XCQ==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-08-08T11:50:09.129957Z"},"content_sha256":"868ef397521f6e9d7ddcfc98d7f0800aed54b416df00bb24a9bf4d91e15f96fa","schema_version":"1.0","event_id":"sha256:868ef397521f6e9d7ddcfc98d7f0800aed54b416df00bb24a9bf4d91e15f96fa"}],"timestamp_proofs":[],"mirror_hints":[{"mirror_type":"https","name":"Pith Resolver","base_url":"https://pith.science","bundle_url":"https://pith.science/pith/NLQU5MFX57FPI7VLZGRLRL34TK/bundle.json","state_url":"https://pith.science/pith/NLQU5MFX57FPI7VLZGRLRL34TK/state.json","well_known_bundle_url":"https://pith.science/.well-known/pith/NLQU5MFX57FPI7VLZGRLRL34TK/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":"2026-08-08T11:50:09Z","links":{"resolver":"https://pith.science/pith/NLQU5MFX57FPI7VLZGRLRL34TK","bundle":"https://pith.science/pith/NLQU5MFX57FPI7VLZGRLRL34TK/bundle.json","state":"https://pith.science/pith/NLQU5MFX57FPI7VLZGRLRL34TK/state.json","well_known_bundle":"https://pith.science/.well-known/pith/NLQU5MFX57FPI7VLZGRLRL34TK/bundle.json"},"state":{"state_type":"pith_open_graph_state","state_version":"1.0","pith_number":"pith:2023:NLQU5MFX57FPI7VLZGRLRL34TK","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":"d3912b24c76222ab8f4b10a7988a347bdef379c07163c7ec74366dafee9dac23","cross_cats_sorted":[],"license":"http://creativecommons.org/licenses/by-nc-nd/4.0/","primary_cat":"cs.CL","submitted_at":"2023-12-06T22:53:34Z","title_canon_sha256":"705b013ebeeeaba997f34a22e95971aaca8510b6af683011b082698f490f4c10"},"schema_version":"1.0","source":{"id":"2312.07551","kind":"arxiv","version":1}},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2312.07551","created_at":"2026-07-05T07:23:32Z"},{"alias_kind":"arxiv_version","alias_value":"2312.07551v1","created_at":"2026-07-05T07:23:32Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2312.07551","created_at":"2026-07-05T07:23:32Z"},{"alias_kind":"pith_short_12","alias_value":"NLQU5MFX57FP","created_at":"2026-07-05T07:23:32Z"},{"alias_kind":"pith_short_16","alias_value":"NLQU5MFX57FPI7VL","created_at":"2026-07-05T07:23:32Z"},{"alias_kind":"pith_short_8","alias_value":"NLQU5MFX","created_at":"2026-07-05T07:23:32Z"}],"graph_snapshots":[{"event_id":"sha256:868ef397521f6e9d7ddcfc98d7f0800aed54b416df00bb24a9bf4d91e15f96fa","target":"graph","created_at":"2026-07-05T07:23:32Z","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/2312.07551/integrity.json","findings":[],"snapshot_sha256":"c28c3603d3b5d939e8dc4c7e95fa8dfce3d595e45f758748cecf8e644a296938","summary":{"advisory":0,"by_detector":{},"critical":0,"informational":0}},"paper":{"abstract_excerpt":"Finetuning language models with reinforcement learning (RL), e.g. from human feedback (HF), is a prominent method for alignment. But optimizing against a reward model can improve on reward while degrading performance in other areas, a phenomenon known as reward hacking, alignment tax, or language drift. First, we argue that commonly-used test metrics are insufficient and instead measure how different algorithms tradeoff between reward and drift. The standard method modified the reward with a Kullback-Lieber (KL) penalty between the online and initial model. We propose Elastic Reset, a new algo","authors_text":"Aaron Courville, Florian Strub, Michael Noukhovitch, Samuel Lavoie","cross_cats":[],"headline":"","license":"http://creativecommons.org/licenses/by-nc-nd/4.0/","primary_cat":"cs.CL","submitted_at":"2023-12-06T22:53:34Z","title":"Language Model Alignment with Elastic Reset"},"references":{"count":0,"internal_anchors":0,"resolved_work":0,"sample":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2312.07551","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":"sha256:212f22b2db378dffe703b9dc7a80a8cf9b761b2fffcc6693c1f46c17800991d8","target":"record","created_at":"2026-07-05T07:23:32Z","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":"d3912b24c76222ab8f4b10a7988a347bdef379c07163c7ec74366dafee9dac23","cross_cats_sorted":[],"license":"http://creativecommons.org/licenses/by-nc-nd/4.0/","primary_cat":"cs.CL","submitted_at":"2023-12-06T22:53:34Z","title_canon_sha256":"705b013ebeeeaba997f34a22e95971aaca8510b6af683011b082698f490f4c10"},"schema_version":"1.0","source":{"id":"2312.07551","kind":"arxiv","version":1}},"canonical_sha256":"6ae14eb0b7efcaf47eabc9a2b8af7c9aacca696ac1dd4cae75c4db95a2a12982","receipt":{"algorithm":"ed25519","builder_version":"pith-number-builder-2026-05-17-v1","canonical_sha256":"6ae14eb0b7efcaf47eabc9a2b8af7c9aacca696ac1dd4cae75c4db95a2a12982","first_computed_at":"2026-07-05T07:23:32.576073Z","key_id":"pith-v1-2026-05","kind":"pith_receipt","last_reissued_at":"2026-07-05T07:23:32.576073Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","receipt_version":"0.3","signature_b64":"uNb9QZjf0FhpcIe06VTXjkH/Q426UfF2C4LuLkb2j2DM9oAXTyZHPhHHXxPcRfZB+MKh3amA2bbliss0aflDBA==","signature_status":"signed_v1","signed_at":"2026-07-05T07:23:32.576591Z","signed_message":"canonical_sha256_bytes"},"source_id":"2312.07551","source_kind":"arxiv","source_version":1}}},"equivocations":[],"invalid_events":[],"applied_event_ids":["sha256:212f22b2db378dffe703b9dc7a80a8cf9b761b2fffcc6693c1f46c17800991d8","sha256:868ef397521f6e9d7ddcfc98d7f0800aed54b416df00bb24a9bf4d91e15f96fa"],"state_sha256":"63f801c35dcedc67a53b5546a1d7749ba1a3d50b4c157725759b8c8f1691bf3f"},"bundle_signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"HMoPHzMFp3Hbx1wvewziK+28l+ROuaqPpYFQFgQGdjIBnLG3TABW+IvsB07mYCv1oKW/tdOxc7mDIQjng9PrAA==","signed_message":"bundle_sha256_bytes","signed_at":"2026-08-08T11:50:09.134405Z","bundle_sha256":"785fb0261050f4e0343448a1c8a047ece70134f1ecee7d241209645c71ecdb53"}}