{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:5TOP3QL74RXK76JBTNLPE77PDN","short_pith_number":"pith:5TOP3QL7","schema_version":"1.0","canonical_sha256":"ecdcfdc17fe46eaff9219b56f27fef1b4e1fec7faffec222d9be669b498f40fd","source":{"kind":"arxiv","id":"2503.09501","version":3},"attestation_state":"computed","paper":{"title":"ReMA: Learning to Meta-think for LLMs with Multi-Agent Reinforcement Learning","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.CL","cs.LG","cs.MA"],"primary_cat":"cs.AI","authors_text":"Hanjing Wang, Jun Wang, Linyi Yang, Mark Schmidt, Shuyue Hu, Weinan Zhang, Xiaoyu Wen, Yan Song, Ying Wen, Yunxiang Li, Ziyu Wan","submitted_at":"2025-03-12T16:05:31Z","abstract_excerpt":"Recent research on Reasoning of Large Language Models (LLMs) has sought to further enhance their performance by integrating meta-thinking -- enabling models to monitor, evaluate, and control their reasoning processes for more adaptive and effective problem-solving. However, current single-agent work lacks a specialized design for acquiring meta-thinking, resulting in low efficacy. To address this challenge, we introduce Reinforced Meta-thinking Agents (ReMA), a novel framework that leverages Multi-Agent Reinforcement Learning (MARL) to elicit meta-thinking behaviors, encouraging LLMs to think "},"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":"2503.09501","kind":"arxiv","version":3},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.AI","submitted_at":"2025-03-12T16:05:31Z","cross_cats_sorted":["cs.CL","cs.LG","cs.MA"],"title_canon_sha256":"b7c0eba3a40500deb0e28334be45be2db8924fe1bbf551e87d8e9ca1d67e294c","abstract_canon_sha256":"c8e09b324596917b2d8f95786197dd905ff11d31e71c6d83bfef9ea73321f458"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:10:06.361782Z","signature_b64":"5aelTVMOwE9RnAzNjKGyQEiFtk5w/vpz8apkTfJ7yUsg40oiM01QkSkOUkqTmvNVxfq6UT9SvDPC6L2ukkkgAg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"ecdcfdc17fe46eaff9219b56f27fef1b4e1fec7faffec222d9be669b498f40fd","last_reissued_at":"2026-07-05T11:10:06.361246Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:10:06.361246Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"ReMA: Learning to Meta-think for LLMs with Multi-Agent Reinforcement Learning","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.CL","cs.LG","cs.MA"],"primary_cat":"cs.AI","authors_text":"Hanjing Wang, Jun Wang, Linyi Yang, Mark Schmidt, Shuyue Hu, Weinan Zhang, Xiaoyu Wen, Yan Song, Ying Wen, Yunxiang Li, Ziyu Wan","submitted_at":"2025-03-12T16:05:31Z","abstract_excerpt":"Recent research on Reasoning of Large Language Models (LLMs) has sought to further enhance their performance by integrating meta-thinking -- enabling models to monitor, evaluate, and control their reasoning processes for more adaptive and effective problem-solving. However, current single-agent work lacks a specialized design for acquiring meta-thinking, resulting in low efficacy. To address this challenge, we introduce Reinforced Meta-thinking Agents (ReMA), a novel framework that leverages Multi-Agent Reinforcement Learning (MARL) to elicit meta-thinking behaviors, encouraging LLMs to think "},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2503.09501","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/2503.09501/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":"2503.09501","created_at":"2026-07-05T11:10:06.361321+00:00"},{"alias_kind":"arxiv_version","alias_value":"2503.09501v3","created_at":"2026-07-05T11:10:06.361321+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2503.09501","created_at":"2026-07-05T11:10:06.361321+00:00"},{"alias_kind":"pith_short_12","alias_value":"5TOP3QL74RXK","created_at":"2026-07-05T11:10:06.361321+00:00"},{"alias_kind":"pith_short_16","alias_value":"5TOP3QL74RXK76JB","created_at":"2026-07-05T11:10:06.361321+00:00"},{"alias_kind":"pith_short_8","alias_value":"5TOP3QL7","created_at":"2026-07-05T11:10:06.361321+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":9,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.26935","citing_title":"Where Do CoT Training Gains Land in LLM based Agents?","ref_index":21,"is_internal_anchor":false},{"citing_arxiv_id":"2504.01990","citing_title":"Advances and Challenges in Foundation Agents: From Brain-Inspired Intelligence to Evolutionary, Collaborative, and Safe Systems","ref_index":169,"is_internal_anchor":false},{"citing_arxiv_id":"2605.21768","citing_title":"Memory-R2: Fair Credit Assignment for Long-Horizon Memory-Augmented LLM Agents","ref_index":17,"is_internal_anchor":false},{"citing_arxiv_id":"2605.18500","citing_title":"Implicit Hierarchical GRPO: Decoupling Tool Invocation from Execution for Tool-Integrated Mathematical Reasoning","ref_index":4,"is_internal_anchor":false},{"citing_arxiv_id":"2605.08401","citing_title":"AIPO: Learning to Reason from Active Interaction","ref_index":64,"is_internal_anchor":false},{"citing_arxiv_id":"2605.08401","citing_title":"AIPO: Learning to Reason from Active Interaction","ref_index":64,"is_internal_anchor":false},{"citing_arxiv_id":"2604.08000","citing_title":"PASK: Toward Intent-Aware Proactive Agents with Long-Term Memory","ref_index":23,"is_internal_anchor":false},{"citing_arxiv_id":"2604.15972","citing_title":"Weak-Link Optimization for Multi-Agent Reasoning and Collaboration","ref_index":34,"is_internal_anchor":false},{"citing_arxiv_id":"2604.20140","citing_title":"HiPO: Hierarchical Preference Optimization for Adaptive Reasoning in LLMs","ref_index":14,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/5TOP3QL74RXK76JBTNLPE77PDN","json":"https://pith.science/pith/5TOP3QL74RXK76JBTNLPE77PDN.json","graph_json":"https://pith.science/api/pith-number/5TOP3QL74RXK76JBTNLPE77PDN/graph.json","events_json":"https://pith.science/api/pith-number/5TOP3QL74RXK76JBTNLPE77PDN/events.json","paper":"https://pith.science/paper/5TOP3QL7"},"agent_actions":{"view_html":"https://pith.science/pith/5TOP3QL74RXK76JBTNLPE77PDN","download_json":"https://pith.science/pith/5TOP3QL74RXK76JBTNLPE77PDN.json","view_paper":"https://pith.science/paper/5TOP3QL7","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2503.09501&json=true","fetch_graph":"https://pith.science/api/pith-number/5TOP3QL74RXK76JBTNLPE77PDN/graph.json","fetch_events":"https://pith.science/api/pith-number/5TOP3QL74RXK76JBTNLPE77PDN/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/5TOP3QL74RXK76JBTNLPE77PDN/action/timestamp_anchor","attest_storage":"https://pith.science/pith/5TOP3QL74RXK76JBTNLPE77PDN/action/storage_attestation","attest_author":"https://pith.science/pith/5TOP3QL74RXK76JBTNLPE77PDN/action/author_attestation","sign_citation":"https://pith.science/pith/5TOP3QL74RXK76JBTNLPE77PDN/action/citation_signature","submit_replication":"https://pith.science/pith/5TOP3QL74RXK76JBTNLPE77PDN/action/replication_record"}},"created_at":"2026-07-05T11:10:06.361321+00:00","updated_at":"2026-07-05T11:10:06.361321+00:00"}