{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:M3PBBCOOUVRH3IOEZVIWJSJN2J","short_pith_number":"pith:M3PBBCOO","schema_version":"1.0","canonical_sha256":"66de1089cea5627da1c4cd5164c92dd25259d9d162f4b400319a9a0e65b66151","source":{"kind":"arxiv","id":"2502.01600","version":3},"attestation_state":"computed","paper":{"title":"Reinforcement Learning for Long-Horizon Interactive LLM Agents","license":"http://creativecommons.org/licenses/by-nc-nd/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.LG","authors_text":"Aleksei Petrenko, Brody Huval, Jackson Hamburger, Kevin Chen, Marco Cusumano-Towner, Philipp Kr\\\"ahenb\\\"uhl, Vladlen Koltun","submitted_at":"2025-02-03T18:35:42Z","abstract_excerpt":"Interactive digital agents (IDAs) leverage APIs of stateful digital environments to perform tasks in response to user requests. While IDAs powered by instruction-tuned large language models (LLMs) can react to feedback from interface invocations in multi-step exchanges, they have not been trained in their respective digital environments. Prior methods accomplish less than half of tasks in sophisticated benchmarks such as AppWorld. We present a reinforcement learning (RL) approach that trains IDAs directly in their target environments. We formalize this training as a partially observable Markov"},"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":"2502.01600","kind":"arxiv","version":3},"metadata":{"license":"http://creativecommons.org/licenses/by-nc-nd/4.0/","primary_cat":"cs.LG","submitted_at":"2025-02-03T18:35:42Z","cross_cats_sorted":["cs.AI"],"title_canon_sha256":"4009082a2c598a4e641ebf2e0ea5d330c4aa6a99b1973b6d954a0ed70be87661","abstract_canon_sha256":"4cd7700f5ecf8cfba416fe85de5b8834e070056a886d5c04b88ecbcaf95ab900"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T10:26:43.514393Z","signature_b64":"qKTi7iEzzSrh7Lpu+bR9fAb2Wkkcaou7p27mLmeGf1NAvjRU3oo3SSHST19FhCbi4PiX4Z/0Y63xjUb9Yj8cBw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"66de1089cea5627da1c4cd5164c92dd25259d9d162f4b400319a9a0e65b66151","last_reissued_at":"2026-07-05T10:26:43.513663Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T10:26:43.513663Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Reinforcement Learning for Long-Horizon Interactive LLM Agents","license":"http://creativecommons.org/licenses/by-nc-nd/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.LG","authors_text":"Aleksei Petrenko, Brody Huval, Jackson Hamburger, Kevin Chen, Marco Cusumano-Towner, Philipp Kr\\\"ahenb\\\"uhl, Vladlen Koltun","submitted_at":"2025-02-03T18:35:42Z","abstract_excerpt":"Interactive digital agents (IDAs) leverage APIs of stateful digital environments to perform tasks in response to user requests. While IDAs powered by instruction-tuned large language models (LLMs) can react to feedback from interface invocations in multi-step exchanges, they have not been trained in their respective digital environments. Prior methods accomplish less than half of tasks in sophisticated benchmarks such as AppWorld. We present a reinforcement learning (RL) approach that trains IDAs directly in their target environments. We formalize this training as a partially observable Markov"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2502.01600","kind":"arxiv","version":3},"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/2502.01600/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":"2502.01600","created_at":"2026-07-05T10:26:43.513748+00:00"},{"alias_kind":"arxiv_version","alias_value":"2502.01600v3","created_at":"2026-07-05T10:26:43.513748+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2502.01600","created_at":"2026-07-05T10:26:43.513748+00:00"},{"alias_kind":"pith_short_12","alias_value":"M3PBBCOOUVRH","created_at":"2026-07-05T10:26:43.513748+00:00"},{"alias_kind":"pith_short_16","alias_value":"M3PBBCOOUVRH3IOE","created_at":"2026-07-05T10:26:43.513748+00:00"},{"alias_kind":"pith_short_8","alias_value":"M3PBBCOO","created_at":"2026-07-05T10:26:43.513748+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":36,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.26918","citing_title":"Diagnosing Task Insensitivity in Language Agents","ref_index":3,"is_internal_anchor":false},{"citing_arxiv_id":"2606.22995","citing_title":"Group-Graph Policy Optimization for Long-Horizon Agentic Reinforcement Learning","ref_index":7,"is_internal_anchor":false},{"citing_arxiv_id":"2606.17735","citing_title":"Shattering the Autoregressive Curse: Dynamic Epistemic Entropy Orchestrated Erasable Reinforcement Learning for LLMs","ref_index":3,"is_internal_anchor":false},{"citing_arxiv_id":"2607.01897","citing_title":"Rank-Then-Act: Reward-Free Control from Frame-Order Progress","ref_index":6,"is_internal_anchor":false},{"citing_arxiv_id":"2607.02431","citing_title":"WorldSample: Closed-loop Real-robot RL with World Modelling","ref_index":3,"is_internal_anchor":false},{"citing_arxiv_id":"2606.08867","citing_title":"Building Customer Support AI Agents at 100M-User Scale: An Evaluation-Driven Framework","ref_index":6,"is_internal_anchor":false},{"citing_arxiv_id":"2606.05885","citing_title":"When Denser Credit Is Not Enough: Evidence-Calibrated Policy Optimization for Long-Horizon LLM Agent Training","ref_index":26,"is_internal_anchor":false},{"citing_arxiv_id":"2606.03108","citing_title":"EvoTrainer: Co-Evolving LLM Policies and Training Harnesses for Autonomous Agentic Reinforcement Learning","ref_index":41,"is_internal_anchor":false},{"citing_arxiv_id":"2606.02355","citing_title":"SIRI: Self-Internalizing Reinforcement Learning with Intrinsic Skills for LLM Agent Training","ref_index":41,"is_internal_anchor":false},{"citing_arxiv_id":"2606.31504","citing_title":"SimpleSearch-VL: A Simple Recipe for Multimodal Agentic Deep Search","ref_index":60,"is_internal_anchor":false},{"citing_arxiv_id":"2605.25384","citing_title":"GeoMathCode: Understanding Interleaved Math-Code Reasoning for Geometry Problem Solving","ref_index":7,"is_internal_anchor":false},{"citing_arxiv_id":"2504.02181","citing_title":"A Survey of Scaling in Large Language Model Reasoning","ref_index":18,"is_internal_anchor":false},{"citing_arxiv_id":"2605.20061","citing_title":"Rewarding Beliefs, Not Actions: Consistency-Guided Credit Assignment for Long-Horizon Agents","ref_index":6,"is_internal_anchor":false},{"citing_arxiv_id":"2605.19762","citing_title":"What Really Improves Mathematical Reasoning: Structured Reasoning Signals Beyond Pure Code","ref_index":5,"is_internal_anchor":false},{"citing_arxiv_id":"2604.26733","citing_title":"FutureWorld: A Live Reinforcement Learning Environment for Predictive Agents with Real-World Outcome Rewards","ref_index":1,"is_internal_anchor":false},{"citing_arxiv_id":"2506.03610","citing_title":"Orak: A Foundational Benchmark for Training and Evaluating LLM Agents on Diverse Video Games","ref_index":49,"is_internal_anchor":false},{"citing_arxiv_id":"2509.24948","citing_title":"World-Env: Leveraging World Model as a Virtual Environment for VLA Post-Training","ref_index":7,"is_internal_anchor":false},{"citing_arxiv_id":"2605.14558","citing_title":"Resolving Action Bottleneck: Agentic Reinforcement Learning Informed by Token-Level Energy","ref_index":3,"is_internal_anchor":false},{"citing_arxiv_id":"2604.03675","citing_title":"OASES: Outcome-Aligned Search-Evaluation Co-Training for Agentic Search","ref_index":1,"is_internal_anchor":false},{"citing_arxiv_id":"2605.06130","citing_title":"Skill1: Unified Evolution of Skill-Augmented Agents via Reinforcement Learning","ref_index":84,"is_internal_anchor":false},{"citing_arxiv_id":"2604.26733","citing_title":"FutureWorld: A Live Reinforcement Learning Environment for Predictive Agents with Real-World Outcome Rewards","ref_index":1,"is_internal_anchor":false},{"citing_arxiv_id":"2604.26733","citing_title":"FutureWorld: A Live Reinforcement Learning Environment for Predictive Agents with Real-World Outcome Rewards","ref_index":1,"is_internal_anchor":false},{"citing_arxiv_id":"2604.24977","citing_title":"A Survey on LLM-based Conversational User Simulation","ref_index":2,"is_internal_anchor":false},{"citing_arxiv_id":"2605.06130","citing_title":"Skill1: Unified Evolution of Skill-Augmented Agents via Reinforcement Learning","ref_index":84,"is_internal_anchor":false},{"citing_arxiv_id":"2605.00425","citing_title":"AEM: Adaptive Entropy Modulation for Multi-Turn Agentic Reinforcement Learning","ref_index":44,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/M3PBBCOOUVRH3IOEZVIWJSJN2J","json":"https://pith.science/pith/M3PBBCOOUVRH3IOEZVIWJSJN2J.json","graph_json":"https://pith.science/api/pith-number/M3PBBCOOUVRH3IOEZVIWJSJN2J/graph.json","events_json":"https://pith.science/api/pith-number/M3PBBCOOUVRH3IOEZVIWJSJN2J/events.json","paper":"https://pith.science/paper/M3PBBCOO"},"agent_actions":{"view_html":"https://pith.science/pith/M3PBBCOOUVRH3IOEZVIWJSJN2J","download_json":"https://pith.science/pith/M3PBBCOOUVRH3IOEZVIWJSJN2J.json","view_paper":"https://pith.science/paper/M3PBBCOO","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2502.01600&json=true","fetch_graph":"https://pith.science/api/pith-number/M3PBBCOOUVRH3IOEZVIWJSJN2J/graph.json","fetch_events":"https://pith.science/api/pith-number/M3PBBCOOUVRH3IOEZVIWJSJN2J/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/M3PBBCOOUVRH3IOEZVIWJSJN2J/action/timestamp_anchor","attest_storage":"https://pith.science/pith/M3PBBCOOUVRH3IOEZVIWJSJN2J/action/storage_attestation","attest_author":"https://pith.science/pith/M3PBBCOOUVRH3IOEZVIWJSJN2J/action/author_attestation","sign_citation":"https://pith.science/pith/M3PBBCOOUVRH3IOEZVIWJSJN2J/action/citation_signature","submit_replication":"https://pith.science/pith/M3PBBCOOUVRH3IOEZVIWJSJN2J/action/replication_record"}},"created_at":"2026-07-05T10:26:43.513748+00:00","updated_at":"2026-07-05T10:26:43.513748+00:00"}