{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:EHL5ZXW5YGNEBAZ23JEZN7LLNC","short_pith_number":"pith:EHL5ZXW5","schema_version":"1.0","canonical_sha256":"21d7dcdeddc19a40833ada4996fd6b68a40dd834216b503dd8034023327826b3","source":{"kind":"arxiv","id":"2504.04736","version":2},"attestation_state":"computed","paper":{"title":"Synthetic Data Generation & Multi-Step RL for Reasoning & Tool Use","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.CL","cs.LG"],"primary_cat":"cs.AI","authors_text":"Anna Goldie, Azalia Mirhoseini, Christopher D. Manning, Hao Zhou, Irene Cai","submitted_at":"2025-04-07T05:20:58Z","abstract_excerpt":"Reinforcement learning has been shown to improve the performance of large language models. However, traditional approaches like RLHF or RLAIF treat the problem as single-step. As focus shifts toward more complex reasoning and agentic tasks, language models must take multiple steps of text generation, reasoning and environment interaction before generating a solution. We propose a synthetic data generation and RL methodology targeting multi-step optimization scenarios. This approach, called Step-Wise Reinforcement Learning (SWiRL), iteratively generates multi-step reasoning and tool use data, 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":"2504.04736","kind":"arxiv","version":2},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.AI","submitted_at":"2025-04-07T05:20:58Z","cross_cats_sorted":["cs.CL","cs.LG"],"title_canon_sha256":"56da1d0d60a1c49baca11beaa0c751678a80bc34e9633644153fe5bfa5299b33","abstract_canon_sha256":"120ef739dce2eddded0d53efed0828c56cc144277d9d049f079eeb893c1327a1"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T10:54:45.891178Z","signature_b64":"lXu0K+9KLFh9+n+tzjgAFRDLcbIvvRKtfRxn2DlHvglpftHe6GXIJ7rUDsSYLN+UHv/IIxegQ4EH8t7J2Yw2Bw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"21d7dcdeddc19a40833ada4996fd6b68a40dd834216b503dd8034023327826b3","last_reissued_at":"2026-07-05T10:54:45.890672Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T10:54:45.890672Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Synthetic Data Generation & Multi-Step RL for Reasoning & Tool Use","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.CL","cs.LG"],"primary_cat":"cs.AI","authors_text":"Anna Goldie, Azalia Mirhoseini, Christopher D. Manning, Hao Zhou, Irene Cai","submitted_at":"2025-04-07T05:20:58Z","abstract_excerpt":"Reinforcement learning has been shown to improve the performance of large language models. However, traditional approaches like RLHF or RLAIF treat the problem as single-step. As focus shifts toward more complex reasoning and agentic tasks, language models must take multiple steps of text generation, reasoning and environment interaction before generating a solution. We propose a synthetic data generation and RL methodology targeting multi-step optimization scenarios. This approach, called Step-Wise Reinforcement Learning (SWiRL), iteratively generates multi-step reasoning and tool use data, a"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2504.04736","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/2504.04736/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":"2504.04736","created_at":"2026-07-05T10:54:45.890734+00:00"},{"alias_kind":"arxiv_version","alias_value":"2504.04736v2","created_at":"2026-07-05T10:54:45.890734+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2504.04736","created_at":"2026-07-05T10:54:45.890734+00:00"},{"alias_kind":"pith_short_12","alias_value":"EHL5ZXW5YGNE","created_at":"2026-07-05T10:54:45.890734+00:00"},{"alias_kind":"pith_short_16","alias_value":"EHL5ZXW5YGNEBAZ2","created_at":"2026-07-05T10:54:45.890734+00:00"},{"alias_kind":"pith_short_8","alias_value":"EHL5ZXW5","created_at":"2026-07-05T10:54:45.890734+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":11,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.02908","citing_title":"WRIT: Write-Read Intensive Trajectory Synthesis for Multi-Turn User-Facing Agents","ref_index":19,"is_internal_anchor":false},{"citing_arxiv_id":"2605.01248","citing_title":"$S^3$-R1: Learning to Retrieve and Answer Step-by-Step with Synthetic Data","ref_index":6,"is_internal_anchor":false},{"citing_arxiv_id":"2606.26122","citing_title":"DocArena: Turning Raw Documents into Controllable Training Environments for Document Search Agents","ref_index":14,"is_internal_anchor":false},{"citing_arxiv_id":"2605.10923","citing_title":"Dynamic Skill Lifecycle Management for Agentic Reinforcement Learning","ref_index":15,"is_internal_anchor":false},{"citing_arxiv_id":"2601.04809","citing_title":"SCALER:Synthetic Scalable Adaptive Learning Environment for Reasoning","ref_index":11,"is_internal_anchor":false},{"citing_arxiv_id":"2508.07407","citing_title":"A Comprehensive Survey of Self-Evolving AI Agents: A New Paradigm Bridging Foundation Models and Lifelong Agentic Systems","ref_index":29,"is_internal_anchor":false},{"citing_arxiv_id":"2504.19678","citing_title":"From LLM Reasoning to Autonomous AI Agents: A Comprehensive Review","ref_index":46,"is_internal_anchor":false},{"citing_arxiv_id":"2605.10923","citing_title":"Dynamic Skill Lifecycle Management for Agentic Reinforcement Learning","ref_index":15,"is_internal_anchor":false},{"citing_arxiv_id":"2604.23333","citing_title":"Process Supervision of Confidence Margin for Calibrated LLM Reasoning","ref_index":19,"is_internal_anchor":false},{"citing_arxiv_id":"2605.01248","citing_title":"$S^3$-R1: Learning to Retrieve and Answer Step-by-Step with Synthetic Data","ref_index":6,"is_internal_anchor":false},{"citing_arxiv_id":"2604.18936","citing_title":"Fine-Tuning Small Reasoning Models for Quantum Field Theory","ref_index":27,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/EHL5ZXW5YGNEBAZ23JEZN7LLNC","json":"https://pith.science/pith/EHL5ZXW5YGNEBAZ23JEZN7LLNC.json","graph_json":"https://pith.science/api/pith-number/EHL5ZXW5YGNEBAZ23JEZN7LLNC/graph.json","events_json":"https://pith.science/api/pith-number/EHL5ZXW5YGNEBAZ23JEZN7LLNC/events.json","paper":"https://pith.science/paper/EHL5ZXW5"},"agent_actions":{"view_html":"https://pith.science/pith/EHL5ZXW5YGNEBAZ23JEZN7LLNC","download_json":"https://pith.science/pith/EHL5ZXW5YGNEBAZ23JEZN7LLNC.json","view_paper":"https://pith.science/paper/EHL5ZXW5","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2504.04736&json=true","fetch_graph":"https://pith.science/api/pith-number/EHL5ZXW5YGNEBAZ23JEZN7LLNC/graph.json","fetch_events":"https://pith.science/api/pith-number/EHL5ZXW5YGNEBAZ23JEZN7LLNC/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/EHL5ZXW5YGNEBAZ23JEZN7LLNC/action/timestamp_anchor","attest_storage":"https://pith.science/pith/EHL5ZXW5YGNEBAZ23JEZN7LLNC/action/storage_attestation","attest_author":"https://pith.science/pith/EHL5ZXW5YGNEBAZ23JEZN7LLNC/action/author_attestation","sign_citation":"https://pith.science/pith/EHL5ZXW5YGNEBAZ23JEZN7LLNC/action/citation_signature","submit_replication":"https://pith.science/pith/EHL5ZXW5YGNEBAZ23JEZN7LLNC/action/replication_record"}},"created_at":"2026-07-05T10:54:45.890734+00:00","updated_at":"2026-07-05T10:54:45.890734+00:00"}