{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:NMPFPV7BKVVNPKG5H2DVQF3IS4","short_pith_number":"pith:NMPFPV7B","schema_version":"1.0","canonical_sha256":"6b1e57d7e1556ad7a8dd3e87581768971f5a488f4876ccd502bf1017fb54d783","source":{"kind":"arxiv","id":"2509.20357","version":2},"attestation_state":"computed","paper":{"title":"Language Models that Think, Chat Better","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Adithya Bhaskar, Danqi Chen, Xi Ye","submitted_at":"2025-09-24T17:57:34Z","abstract_excerpt":"Reinforcement learning with verifiable rewards (RLVR) trains language models to use long chain-of-thought reasoning (CoT) in domains like mathematics and code with rule-based verifiers. However, long CoT learned through RLVR does not generalize well to open-ended tasks -- such as writing essay outlines or making meal plans -- where humans reason routinely. This paper establishes the benefits of long CoT for general-purpose chat capabilities and introduces RL with Model-rewarded Thinking (RLMT)1, which pushes RLVR beyond verifiable domains. Using diverse real-world prompts, RLMT requires LMs to"},"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":"2509.20357","kind":"arxiv","version":2},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CL","submitted_at":"2025-09-24T17:57:34Z","cross_cats_sorted":[],"title_canon_sha256":"851157b2eb5242c3d4ddb3dd5d3cb9ef5650c6042d09e0252c81ae75249b5ffa","abstract_canon_sha256":"1d2e98e96454addd06cd055db3d84359a42d5af1b1f9df65ecaaaf811c0e2d14"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-08-18T01:14:00.852840Z","signature_b64":"CZVIms3O7papaanh/JM2jhhvi/rFzmrEZtr0/frfXY2pzZmAA3r8p2t3fchjVio267Mgv/RbkSMHfX+VPt4GCQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"6b1e57d7e1556ad7a8dd3e87581768971f5a488f4876ccd502bf1017fb54d783","last_reissued_at":"2026-08-18T01:14:00.851120Z","signature_status":"signed_v1","first_computed_at":"2026-08-18T01:14:00.851120Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Language Models that Think, Chat Better","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Adithya Bhaskar, Danqi Chen, Xi Ye","submitted_at":"2025-09-24T17:57:34Z","abstract_excerpt":"Reinforcement learning with verifiable rewards (RLVR) trains language models to use long chain-of-thought reasoning (CoT) in domains like mathematics and code with rule-based verifiers. However, long CoT learned through RLVR does not generalize well to open-ended tasks -- such as writing essay outlines or making meal plans -- where humans reason routinely. This paper establishes the benefits of long CoT for general-purpose chat capabilities and introduces RL with Model-rewarded Thinking (RLMT)1, which pushes RLVR beyond verifiable domains. Using diverse real-world prompts, RLMT requires LMs to"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2509.20357","kind":"arxiv","version":2},"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/2509.20357/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":"2509.20357","created_at":"2026-08-18T01:14:00.851736+00:00"},{"alias_kind":"arxiv_version","alias_value":"2509.20357v2","created_at":"2026-08-18T01:14:00.851736+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2509.20357","created_at":"2026-08-18T01:14:00.851736+00:00"},{"alias_kind":"pith_short_12","alias_value":"NMPFPV7BKVVN","created_at":"2026-08-18T01:14:00.851736+00:00"},{"alias_kind":"pith_short_16","alias_value":"NMPFPV7BKVVNPKG5","created_at":"2026-08-18T01:14:00.851736+00:00"},{"alias_kind":"pith_short_8","alias_value":"NMPFPV7B","created_at":"2026-08-18T01:14:00.851736+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":7,"internal_anchor_count":7,"sample":[{"citing_arxiv_id":"2606.30989","citing_title":"Wait, am I Being Fair? Characterizing Deductive Stereotyping and Mitigating It with Fair-GCG","ref_index":96,"is_internal_anchor":true},{"citing_arxiv_id":"2606.00609","citing_title":"CARE-RL: Capability-Aware Reinforcement Learning for Mitigating Cross-Domain Conflicts","ref_index":37,"is_internal_anchor":true},{"citing_arxiv_id":"2606.00609","citing_title":"CARE-RL: Capability-Aware Reinforcement Learning for Mitigating Cross-Domain Conflicts","ref_index":53,"is_internal_anchor":true},{"citing_arxiv_id":"2509.08827","citing_title":"A Survey of Reinforcement Learning for Large Reasoning Models","ref_index":36,"is_internal_anchor":true},{"citing_arxiv_id":"2511.09907","citing_title":"Learning to Pose Problems: Reasoning-Driven and Solver-Adaptive Data Synthesis","ref_index":2,"is_internal_anchor":true},{"citing_arxiv_id":"2603.15646","citing_title":"Alternating Reinforcement Learning with Contextual Rubric Rewards: Beyond the Scalarization Strategy","ref_index":5,"is_internal_anchor":true},{"citing_arxiv_id":"2604.08477","citing_title":"SUPERNOVA: Eliciting General Reasoning in LLMs with Reinforcement Learning on Natural Instructions","ref_index":4,"is_internal_anchor":true}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/NMPFPV7BKVVNPKG5H2DVQF3IS4","json":"https://pith.science/pith/NMPFPV7BKVVNPKG5H2DVQF3IS4.json","graph_json":"https://pith.science/api/pith-number/NMPFPV7BKVVNPKG5H2DVQF3IS4/graph.json","events_json":"https://pith.science/api/pith-number/NMPFPV7BKVVNPKG5H2DVQF3IS4/events.json","paper":"https://pith.science/paper/NMPFPV7B"},"agent_actions":{"view_html":"https://pith.science/pith/NMPFPV7BKVVNPKG5H2DVQF3IS4","download_json":"https://pith.science/pith/NMPFPV7BKVVNPKG5H2DVQF3IS4.json","view_paper":"https://pith.science/paper/NMPFPV7B","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2509.20357&json=true","fetch_graph":"https://pith.science/api/pith-number/NMPFPV7BKVVNPKG5H2DVQF3IS4/graph.json","fetch_events":"https://pith.science/api/pith-number/NMPFPV7BKVVNPKG5H2DVQF3IS4/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/NMPFPV7BKVVNPKG5H2DVQF3IS4/action/timestamp_anchor","attest_storage":"https://pith.science/pith/NMPFPV7BKVVNPKG5H2DVQF3IS4/action/storage_attestation","attest_author":"https://pith.science/pith/NMPFPV7BKVVNPKG5H2DVQF3IS4/action/author_attestation","sign_citation":"https://pith.science/pith/NMPFPV7BKVVNPKG5H2DVQF3IS4/action/citation_signature","submit_replication":"https://pith.science/pith/NMPFPV7BKVVNPKG5H2DVQF3IS4/action/replication_record"}},"created_at":"2026-08-18T01:14:00.851736+00:00","updated_at":"2026-08-18T01:14:00.851736+00:00"}