{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:MR6TTSE64MCEPVSQ5JGGFXXQIM","short_pith_number":"pith:MR6TTSE6","schema_version":"1.0","canonical_sha256":"647d39c89ee30447d650ea4c62def0433db3dcf64edb2b7b54cd3821bf7794d8","source":{"kind":"arxiv","id":"2505.11081","version":1},"attestation_state":"computed","paper":{"title":"ShiQ: Bringing back Bellman to LLMs","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Eugene Tarassov, Florian Strub, Irem Ergun, Matthieu Geist, Nathan Grinsztajn, Olivier Pietquin, Omar D. Domingues, Pierre Clavier, Pierre H. Richemond, Raphael Avalos, Yannis Flet-Berliac","submitted_at":"2025-05-16T10:12:11Z","abstract_excerpt":"The fine-tuning of pre-trained large language models (LLMs) using reinforcement learning (RL) is generally formulated as direct policy optimization. This approach was naturally favored as it efficiently improves a pretrained LLM, seen as an initial policy. Another RL paradigm, Q-learning methods, has received far less attention in the LLM community while demonstrating major success in various non-LLM RL tasks. In particular, Q-learning effectiveness comes from its sample efficiency and ability to learn offline, which is particularly valuable given the high computational cost of sampling with L"},"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":"2505.11081","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2025-05-16T10:12:11Z","cross_cats_sorted":[],"title_canon_sha256":"3699f4342aaaee7615122fe223cfbdcdd3697eab7c8ea293a44986899e52180d","abstract_canon_sha256":"1156f7e448a746f6fffe44615fdef48a3f0aa83131175d4e37b198ded790bc71"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:04:10.472840Z","signature_b64":"Bo1i0G0da6wiBjskVQWCfogNdbWStNfFO1J5b7pe5TlBmhVU/mZt4bzLzjE1Axef+1gW1bnAEJ1XFufP5Z2NCw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"647d39c89ee30447d650ea4c62def0433db3dcf64edb2b7b54cd3821bf7794d8","last_reissued_at":"2026-07-05T11:04:10.472296Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:04:10.472296Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"ShiQ: Bringing back Bellman to LLMs","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Eugene Tarassov, Florian Strub, Irem Ergun, Matthieu Geist, Nathan Grinsztajn, Olivier Pietquin, Omar D. Domingues, Pierre Clavier, Pierre H. Richemond, Raphael Avalos, Yannis Flet-Berliac","submitted_at":"2025-05-16T10:12:11Z","abstract_excerpt":"The fine-tuning of pre-trained large language models (LLMs) using reinforcement learning (RL) is generally formulated as direct policy optimization. This approach was naturally favored as it efficiently improves a pretrained LLM, seen as an initial policy. Another RL paradigm, Q-learning methods, has received far less attention in the LLM community while demonstrating major success in various non-LLM RL tasks. In particular, Q-learning effectiveness comes from its sample efficiency and ability to learn offline, which is particularly valuable given the high computational cost of sampling with L"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2505.11081","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/2505.11081/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":"2505.11081","created_at":"2026-07-05T11:04:10.472371+00:00"},{"alias_kind":"arxiv_version","alias_value":"2505.11081v1","created_at":"2026-07-05T11:04:10.472371+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2505.11081","created_at":"2026-07-05T11:04:10.472371+00:00"},{"alias_kind":"pith_short_12","alias_value":"MR6TTSE64MCE","created_at":"2026-07-05T11:04:10.472371+00:00"},{"alias_kind":"pith_short_16","alias_value":"MR6TTSE64MCEPVSQ","created_at":"2026-07-05T11:04:10.472371+00:00"},{"alias_kind":"pith_short_8","alias_value":"MR6TTSE6","created_at":"2026-07-05T11:04:10.472371+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":2,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.21943","citing_title":"Modularized Reinforcement Learning on LLMs: From MDP Creation to Exploration and Learning","ref_index":33,"is_internal_anchor":false},{"citing_arxiv_id":"2512.15605","citing_title":"Autoregressive Language Models are Secretly Energy-Based Models: Insights into the Lookahead Capabilities of Next-Token Prediction","ref_index":11,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/MR6TTSE64MCEPVSQ5JGGFXXQIM","json":"https://pith.science/pith/MR6TTSE64MCEPVSQ5JGGFXXQIM.json","graph_json":"https://pith.science/api/pith-number/MR6TTSE64MCEPVSQ5JGGFXXQIM/graph.json","events_json":"https://pith.science/api/pith-number/MR6TTSE64MCEPVSQ5JGGFXXQIM/events.json","paper":"https://pith.science/paper/MR6TTSE6"},"agent_actions":{"view_html":"https://pith.science/pith/MR6TTSE64MCEPVSQ5JGGFXXQIM","download_json":"https://pith.science/pith/MR6TTSE64MCEPVSQ5JGGFXXQIM.json","view_paper":"https://pith.science/paper/MR6TTSE6","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2505.11081&json=true","fetch_graph":"https://pith.science/api/pith-number/MR6TTSE64MCEPVSQ5JGGFXXQIM/graph.json","fetch_events":"https://pith.science/api/pith-number/MR6TTSE64MCEPVSQ5JGGFXXQIM/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/MR6TTSE64MCEPVSQ5JGGFXXQIM/action/timestamp_anchor","attest_storage":"https://pith.science/pith/MR6TTSE64MCEPVSQ5JGGFXXQIM/action/storage_attestation","attest_author":"https://pith.science/pith/MR6TTSE64MCEPVSQ5JGGFXXQIM/action/author_attestation","sign_citation":"https://pith.science/pith/MR6TTSE64MCEPVSQ5JGGFXXQIM/action/citation_signature","submit_replication":"https://pith.science/pith/MR6TTSE64MCEPVSQ5JGGFXXQIM/action/replication_record"}},"created_at":"2026-07-05T11:04:10.472371+00:00","updated_at":"2026-07-05T11:04:10.472371+00:00"}