{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:4HPHFHH5UYY7QTGPI6WFBJOIBD","short_pith_number":"pith:4HPHFHH5","schema_version":"1.0","canonical_sha256":"e1de729cfda631f84ccf47ac50a5c808d0039710ff1634fba3a864d54491274d","source":{"kind":"arxiv","id":"2605.15464","version":1},"attestation_state":"computed","paper":{"title":"GRLO: Towards Generalizable Reinforcement Learning in Open-Ended Environments from Zero","license":"http://creativecommons.org/licenses/by/4.0/","headline":"Reinforcement learning from open-ended conversations transfers to improve math and code performance without domain-specific training.","cross_cats":["cs.AI","cs.CL"],"primary_cat":"cs.LG","authors_text":"Shangjian Yin, Yue Dong, Yu Fu, Zhouxing Shi","submitted_at":"2026-05-14T23:05:23Z","abstract_excerpt":"Post-training has become a crucial step for unlocking the capabilities of large language models, with reinforcement learning (RL) emerging as a critical paradigm. Recent RL-based post-training has increasingly split into two paradigms: reinforcement learning from human feedback (RLHF), which optimizes models using human preference signals in target domains, and reinforcement learning from verifiable rewards (RLVR), which operates in verifier-backed environments. The latter has dominated recent reasoning-oriented post-training because it delivers stronger gains and higher efficiency on domain-s"},"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":true,"formal_links_present":true},"canonical_record":{"source":{"id":"2605.15464","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2026-05-14T23:05:23Z","cross_cats_sorted":["cs.AI","cs.CL"],"title_canon_sha256":"05b8f09e958e40042d9785d856788815ef5a80527455c9032cbd99bfc43a9a7e","abstract_canon_sha256":"5024d189c952fe99f92d232998f50195460df206b26b16ebea862bea00ae5fb2"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-05-20T00:00:59.903621Z","signature_b64":"XuaF5qQpGYB2XGGxbMAUug0OCuSxqm154Vfx5lgUgMt2+kpnHUG9KTL2Tl4fIKjUv2mbYs/bi3IvNIkD2g8KBw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"e1de729cfda631f84ccf47ac50a5c808d0039710ff1634fba3a864d54491274d","last_reissued_at":"2026-05-20T00:00:59.902791Z","signature_status":"signed_v1","first_computed_at":"2026-05-20T00:00:59.902791Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"GRLO: Towards Generalizable Reinforcement Learning in Open-Ended Environments from Zero","license":"http://creativecommons.org/licenses/by/4.0/","headline":"Reinforcement learning from open-ended conversations transfers to improve math and code performance without domain-specific training.","cross_cats":["cs.AI","cs.CL"],"primary_cat":"cs.LG","authors_text":"Shangjian Yin, Yue Dong, Yu Fu, Zhouxing Shi","submitted_at":"2026-05-14T23:05:23Z","abstract_excerpt":"Post-training has become a crucial step for unlocking the capabilities of large language models, with reinforcement learning (RL) emerging as a critical paradigm. Recent RL-based post-training has increasingly split into two paradigms: reinforcement learning from human feedback (RLHF), which optimizes models using human preference signals in target domains, and reinforcement learning from verifiable rewards (RLVR), which operates in verifier-backed environments. The latter has dominated recent reasoning-oriented post-training because it delivers stronger gains and higher efficiency on domain-s"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"on Qwen3-4B-Base backbone, GRLO improves the average performance across all domains from 24.1 to 63.1 with only 5K prompts and 22.7 GPU hours, requiring about 46× less data and 68× less compute than a strong in-domain RLVR baseline. The resulting model is even competitive with Qwen's released post-trained models which required a much larger training cost.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"The assumption that conversational abilities explicitly acquired through RLHF in open-ended environments will implicitly transfer to downstream tasks such as mathematical reasoning and code generation without any direct training on those domains.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"GRLO shows RLHF from scratch on 5K open-ended prompts raises average performance from 24.1 to 63.1 across domains on Qwen3-4B-Base using 46x less data and 68x less compute than in-domain RLVR while remaining competitive with heavily post-trained models.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"Reinforcement learning from open-ended conversations transfers to improve math and code performance without domain-specific training.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"b52cc292a28ffbb57476ea2b5bd30540929551194fdfe7047134831080399668"},"source":{"id":"2605.15464","kind":"arxiv","version":1},"verdict":{"id":"0215fb6d-30f0-4fc4-b98b-9a8a3bdc08f1","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-19T15:09:52.512569Z","strongest_claim":"on Qwen3-4B-Base backbone, GRLO improves the average performance across all domains from 24.1 to 63.1 with only 5K prompts and 22.7 GPU hours, requiring about 46× less data and 68× less compute than a strong in-domain RLVR baseline. The resulting model is even competitive with Qwen's released post-trained models which required a much larger training cost.","one_line_summary":"GRLO shows RLHF from scratch on 5K open-ended prompts raises average performance from 24.1 to 63.1 across domains on Qwen3-4B-Base using 46x less data and 68x less compute than in-domain RLVR while remaining competitive with heavily post-trained models.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"The assumption that conversational abilities explicitly acquired through RLHF in open-ended environments will implicitly transfer to downstream tasks such as mathematical reasoning and code generation without any direct training on those domains.","pith_extraction_headline":"Reinforcement learning from open-ended conversations transfers to improve math and code performance without domain-specific training."},"integrity":{"clean":true,"summary":{"advisory":0,"critical":0,"by_detector":{},"informational":0},"endpoint":"/pith/2605.15464/integrity.json","findings":[],"available":true,"detectors_run":[{"name":"citation_quote_validity","ran_at":"2026-05-19T15:49:41.857043Z","status":"completed","version":"0.1.0","findings_count":0},{"name":"doi_title_agreement","ran_at":"2026-05-19T15:31:17.724755Z","status":"completed","version":"1.0.0","findings_count":0},{"name":"cited_work_retraction","ran_at":"2026-05-19T15:22:30.286168Z","status":"completed","version":"1.0.0","findings_count":0},{"name":"doi_compliance","ran_at":"2026-05-19T15:20:47.199183Z","status":"completed","version":"1.0.0","findings_count":0},{"name":"claim_evidence","ran_at":"2026-05-19T14:21:54.094801Z","status":"completed","version":"1.0.0","findings_count":0},{"name":"ai_meta_artifact","ran_at":"2026-05-19T13:33:22.668172Z","status":"skipped","version":"1.0.0","findings_count":0}],"snapshot_sha256":"63ef73283a179f664152fa94d2578607b178ecc36da8fd0e1558a49309d0b384"},"references":{"count":45,"sample":[{"doi":"","year":2017,"title":"Proximal Policy Optimization Algorithms , author=. 2017 , eprint=","work_id":"5a794cfe-c0fb-4191-a347-b4e935d39d00","ref_index":1,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2024,"title":"2024 , journal =","work_id":"a27bdbdd-8dfd-46d9-a53f-6413aeedb275","ref_index":2,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2025,"title":"Language Models that Think, Chat Better , author=. 2025 , eprint=","work_id":"fa3ba551-f1c0-46c6-b497-ae8cb91692db","ref_index":3,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2025,"title":"Long-Short Chain-of-Thought Mixture Supervised Fine-Tuning Eliciting Efficient Reasoning in Large Language Models , author=. 2025 , eprint=","work_id":"a4d0aef2-a48c-4084-a90e-3fbff10046dc","ref_index":4,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2025,"title":"Length-Controlled AlpacaEval: A Simple Way to Debias Automatic Evaluators , author=. 2025 , eprint=","work_id":"c0cf9815-749f-4dc1-8de3-a97c3085a69d","ref_index":5,"cited_arxiv_id":"","is_internal_anchor":false}],"resolved_work":45,"snapshot_sha256":"1b300308b5ddbd9fc93acb0a712d7b9e17879e2d4a05da60f410d064305b70d3","internal_anchors":3},"formal_canon":{"evidence_count":2,"snapshot_sha256":"5d126ce1dffc82271ace75c99e14db57b8425710daf83700e13265df4b8a1fd4"},"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":"2605.15464","created_at":"2026-05-20T00:00:59.902917+00:00"},{"alias_kind":"arxiv_version","alias_value":"2605.15464v1","created_at":"2026-05-20T00:00:59.902917+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2605.15464","created_at":"2026-05-20T00:00:59.902917+00:00"},{"alias_kind":"pith_short_12","alias_value":"4HPHFHH5UYY7","created_at":"2026-05-20T00:00:59.902917+00:00"},{"alias_kind":"pith_short_16","alias_value":"4HPHFHH5UYY7QTGP","created_at":"2026-05-20T00:00:59.902917+00:00"},{"alias_kind":"pith_short_8","alias_value":"4HPHFHH5","created_at":"2026-05-20T00:00:59.902917+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":0,"internal_anchor_count":0,"sample":[]},"formal_canon":{"evidence_count":2,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/4HPHFHH5UYY7QTGPI6WFBJOIBD","json":"https://pith.science/pith/4HPHFHH5UYY7QTGPI6WFBJOIBD.json","graph_json":"https://pith.science/api/pith-number/4HPHFHH5UYY7QTGPI6WFBJOIBD/graph.json","events_json":"https://pith.science/api/pith-number/4HPHFHH5UYY7QTGPI6WFBJOIBD/events.json","paper":"https://pith.science/paper/4HPHFHH5"},"agent_actions":{"view_html":"https://pith.science/pith/4HPHFHH5UYY7QTGPI6WFBJOIBD","download_json":"https://pith.science/pith/4HPHFHH5UYY7QTGPI6WFBJOIBD.json","view_paper":"https://pith.science/paper/4HPHFHH5","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2605.15464&json=true","fetch_graph":"https://pith.science/api/pith-number/4HPHFHH5UYY7QTGPI6WFBJOIBD/graph.json","fetch_events":"https://pith.science/api/pith-number/4HPHFHH5UYY7QTGPI6WFBJOIBD/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/4HPHFHH5UYY7QTGPI6WFBJOIBD/action/timestamp_anchor","attest_storage":"https://pith.science/pith/4HPHFHH5UYY7QTGPI6WFBJOIBD/action/storage_attestation","attest_author":"https://pith.science/pith/4HPHFHH5UYY7QTGPI6WFBJOIBD/action/author_attestation","sign_citation":"https://pith.science/pith/4HPHFHH5UYY7QTGPI6WFBJOIBD/action/citation_signature","submit_replication":"https://pith.science/pith/4HPHFHH5UYY7QTGPI6WFBJOIBD/action/replication_record"}},"created_at":"2026-05-20T00:00:59.902917+00:00","updated_at":"2026-05-20T00:00:59.902917+00:00"}