{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:HSYGU7KMA2FLLY7IW27AG3LMB3","short_pith_number":"pith:HSYGU7KM","schema_version":"1.0","canonical_sha256":"3cb06a7d4c068ab5e3e8b6be036d6c0ecac88784ea6c6868940724411b5dd7a0","source":{"kind":"arxiv","id":"2605.27701","version":1},"attestation_state":"computed","paper":{"title":"Cross-Entropy Games and Frost Training","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.AI","authors_text":"Arthur Renard, Cl\\'ement Hongler, Franck Gabriel, Valentin Hartmann","submitted_at":"2026-05-26T21:20:45Z","abstract_excerpt":"We present Frost Training, a method for improving Monte Carlo-based policy optimization for a large family of LLM-as-a-judge tasks called Cross-Entropy Games. The key idea is to exploit the gradient of the reward function in embedding space. This signal is used in the Greedy Coordinate Gradient (GCG) jailbreaking technique; we demonstrate for the first time that it can also be used to boost model training. We validate our method using GRPO training for maximum-likelihood infilling. Frost Training improves the model's ability to generate high-scoring outputs, reaching higher maximum scores in a"},"verification_status":{"content_addressed":true,"pith_receipt":true,"author_attested":false,"weak_author_claims":0,"strong_author_claims":0,"externally_anchored":false,"storage_verified":false,"citation_signatures":0,"replication_records":0,"graph_snapshot":true,"references_resolved":false,"formal_links_present":false},"canonical_record":{"source":{"id":"2605.27701","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.AI","submitted_at":"2026-05-26T21:20:45Z","cross_cats_sorted":[],"title_canon_sha256":"bf51cb67026d59770f62d9efda880571735e27303b3204cb7aa877dcb479c0a4","abstract_canon_sha256":"ce3b2129f3a62243045f6afd7e3cba44eaff72956ecda5639d39a3bd05447aa6"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-05-28T01:04:46.622647Z","signature_b64":"SEpHUEO1S22BZ7vVhmql387FnDBbbhSeKLVzOLpVpnwVG+knxwQXN9qENqaVbZnNxULYCCD7R8NWYbX5gftuDg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"3cb06a7d4c068ab5e3e8b6be036d6c0ecac88784ea6c6868940724411b5dd7a0","last_reissued_at":"2026-05-28T01:04:46.622229Z","signature_status":"signed_v1","first_computed_at":"2026-05-28T01:04:46.622229Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Cross-Entropy Games and Frost Training","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.AI","authors_text":"Arthur Renard, Cl\\'ement Hongler, Franck Gabriel, Valentin Hartmann","submitted_at":"2026-05-26T21:20:45Z","abstract_excerpt":"We present Frost Training, a method for improving Monte Carlo-based policy optimization for a large family of LLM-as-a-judge tasks called Cross-Entropy Games. The key idea is to exploit the gradient of the reward function in embedding space. This signal is used in the Greedy Coordinate Gradient (GCG) jailbreaking technique; we demonstrate for the first time that it can also be used to boost model training. We validate our method using GRPO training for maximum-likelihood infilling. Frost Training improves the model's ability to generate high-scoring outputs, reaching higher maximum scores in a"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2605.27701","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/2605.27701/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"},"aliases":[{"alias_kind":"arxiv","alias_value":"2605.27701","created_at":"2026-05-28T01:04:46.622286+00:00"},{"alias_kind":"arxiv_version","alias_value":"2605.27701v1","created_at":"2026-05-28T01:04:46.622286+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2605.27701","created_at":"2026-05-28T01:04:46.622286+00:00"},{"alias_kind":"pith_short_12","alias_value":"HSYGU7KMA2FL","created_at":"2026-05-28T01:04:46.622286+00:00"},{"alias_kind":"pith_short_16","alias_value":"HSYGU7KMA2FLLY7I","created_at":"2026-05-28T01:04:46.622286+00:00"},{"alias_kind":"pith_short_8","alias_value":"HSYGU7KM","created_at":"2026-05-28T01:04:46.622286+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":0,"internal_anchor_count":0,"sample":[]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/HSYGU7KMA2FLLY7IW27AG3LMB3","json":"https://pith.science/pith/HSYGU7KMA2FLLY7IW27AG3LMB3.json","graph_json":"https://pith.science/api/pith-number/HSYGU7KMA2FLLY7IW27AG3LMB3/graph.json","events_json":"https://pith.science/api/pith-number/HSYGU7KMA2FLLY7IW27AG3LMB3/events.json","paper":"https://pith.science/paper/HSYGU7KM"},"agent_actions":{"view_html":"https://pith.science/pith/HSYGU7KMA2FLLY7IW27AG3LMB3","download_json":"https://pith.science/pith/HSYGU7KMA2FLLY7IW27AG3LMB3.json","view_paper":"https://pith.science/paper/HSYGU7KM","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2605.27701&json=true","fetch_graph":"https://pith.science/api/pith-number/HSYGU7KMA2FLLY7IW27AG3LMB3/graph.json","fetch_events":"https://pith.science/api/pith-number/HSYGU7KMA2FLLY7IW27AG3LMB3/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/HSYGU7KMA2FLLY7IW27AG3LMB3/action/timestamp_anchor","attest_storage":"https://pith.science/pith/HSYGU7KMA2FLLY7IW27AG3LMB3/action/storage_attestation","attest_author":"https://pith.science/pith/HSYGU7KMA2FLLY7IW27AG3LMB3/action/author_attestation","sign_citation":"https://pith.science/pith/HSYGU7KMA2FLLY7IW27AG3LMB3/action/citation_signature","submit_replication":"https://pith.science/pith/HSYGU7KMA2FLLY7IW27AG3LMB3/action/replication_record"}},"created_at":"2026-05-28T01:04:46.622286+00:00","updated_at":"2026-05-28T01:04:46.622286+00:00"}