{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2023:656DEUNACXYL43CYGYJA4Y6JXK","short_pith_number":"pith:656DEUNA","schema_version":"1.0","canonical_sha256":"f77c3251a015f0be6c5836120e63c9ba9c42f0af9f480f7a1fb7885929e16031","source":{"kind":"arxiv","id":"2305.19118","version":4},"attestation_state":"computed","paper":{"title":"Encouraging Divergent Thinking in Large Language Models through Multi-Agent Debate","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"Large language models overcome stuck reasoning by having multiple agents argue tit-for-tat under a judge instead of reflecting alone.","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Rui Wang, Shuming Shi, Tian Liang, Wenxiang Jiao, Xing Wang, Yan Wang, Yujiu Yang, Zhaopeng Tu, Zhiwei He","submitted_at":"2023-05-30T15:25:45Z","abstract_excerpt":"Modern large language models (LLMs) like ChatGPT have shown remarkable performance on general language tasks but still struggle on complex reasoning tasks, which drives the research on cognitive behaviors of LLMs to explore human-like problem-solving strategies. Along this direction, one representative strategy is self-reflection, which asks an LLM to refine the solution with the feedback generated by itself iteratively. However, our study shows that such reflection-style methods suffer from the Degeneration-of-Thought (DoT) problem: once the LLM has established confidence in its solutions, it"},"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":"2305.19118","kind":"arxiv","version":4},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CL","submitted_at":"2023-05-30T15:25:45Z","cross_cats_sorted":[],"title_canon_sha256":"314e98a7f73c2b0c1bc6afb2a80a822404b6a592830fef72602665def411d52f","abstract_canon_sha256":"536774355440d16d32852d1d5b7dbfec57261204744c37a7e2a685e9e8d8360d"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-05-18T03:30:52.355108Z","signature_b64":"yV38JiDuAZnAgLkNkNuriNZI+eXHgf0jLx9sfXIMUZcF0djnuC5yj2XzUrfroez2kYw+trCDf7UJkABzsKHcBg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"f77c3251a015f0be6c5836120e63c9ba9c42f0af9f480f7a1fb7885929e16031","last_reissued_at":"2026-05-18T03:30:52.354097Z","signature_status":"signed_v1","first_computed_at":"2026-05-18T03:30:52.354097Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Encouraging Divergent Thinking in Large Language Models through Multi-Agent Debate","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"Large language models overcome stuck reasoning by having multiple agents argue tit-for-tat under a judge instead of reflecting alone.","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Rui Wang, Shuming Shi, Tian Liang, Wenxiang Jiao, Xing Wang, Yan Wang, Yujiu Yang, Zhaopeng Tu, Zhiwei He","submitted_at":"2023-05-30T15:25:45Z","abstract_excerpt":"Modern large language models (LLMs) like ChatGPT have shown remarkable performance on general language tasks but still struggle on complex reasoning tasks, which drives the research on cognitive behaviors of LLMs to explore human-like problem-solving strategies. Along this direction, one representative strategy is self-reflection, which asks an LLM to refine the solution with the feedback generated by itself iteratively. However, our study shows that such reflection-style methods suffer from the Degeneration-of-Thought (DoT) problem: once the LLM has established confidence in its solutions, it"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"Our MAD framework encourages divergent thinking in LLMs which would be helpful for tasks that require deep levels of contemplation. Experiment results on two challenging datasets, commonsense machine translation and counter-intuitive arithmetic reasoning, demonstrate the effectiveness of our MAD framework.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"That a judge LLM can fairly evaluate and synthesize the debate without inheriting the same Degeneration-of-Thought bias, and that the tit-for-tat dynamic reliably produces novel thoughts rather than escalating errors.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"Multi-agent debate with tit-for-tat arguments and a judge LLM improves reasoning by preventing LLMs from locking into incorrect initial solutions.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"Large language models overcome stuck reasoning by having multiple agents argue tit-for-tat under a judge instead of reflecting alone.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"085674e93ac54155e332f700862953f255318eb6d06e2bb45bb3b7918b72a98c"},"source":{"id":"2305.19118","kind":"arxiv","version":4},"verdict":{"id":"dfd64283-d483-4049-bf8e-c6e99c890a83","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-13T23:56:00.948856Z","strongest_claim":"Our MAD framework encourages divergent thinking in LLMs which would be helpful for tasks that require deep levels of contemplation. Experiment results on two challenging datasets, commonsense machine translation and counter-intuitive arithmetic reasoning, demonstrate the effectiveness of our MAD framework.","one_line_summary":"Multi-agent debate with tit-for-tat arguments and a judge LLM improves reasoning by preventing LLMs from locking into incorrect initial solutions.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"That a judge LLM can fairly evaluate and synthesize the debate without inheriting the same Degeneration-of-Thought bias, and that the tit-for-tat dynamic reliably produces novel thoughts rather than escalating errors.","pith_extraction_headline":"Large language models overcome stuck reasoning by having multiple agents argue tit-for-tat under a judge instead of reflecting alone."},"references":{"count":286,"sample":[{"doi":"","year":2023,"title":"Proceedings of the 2023 Conference on Empirical Methods in Natural Language Processing , pages=","work_id":"2bd92650-b903-4f69-9ae1-8c4085db98a8","ref_index":1,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":null,"title":"A Multitask, Multilingual, Multimodal Evaluation of ChatGPT on Reasoning, Hallucination, and Interactivity , author=. Proceedings of the 13th International Joint Conference on Natural Language Process","work_id":"ceda3664-76be-4173-ac42-7b31fdf7124c","ref_index":3,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2015,"title":"Proceedings of the 2015 Conference on Empirical Methods in Natural Language Processing , pages=","work_id":"2232ea1d-9981-4dc7-8380-6b7c6e81f564","ref_index":7,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":null,"title":"Proceedings of the 36th Annual ACM Symposium on User Interface Software and Technology , pages=","work_id":"cf0f4dfd-48c8-4bea-b512-f8814deee1fd","ref_index":12,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":null,"title":"Advances in neural information processing systems , volume=","work_id":"f3923cef-fa32-49c8-8280-9c6bb5d18d27","ref_index":13,"cited_arxiv_id":"","is_internal_anchor":false}],"resolved_work":286,"snapshot_sha256":"856cc4182993b9ff6f11949be4c4f8fdff74f8f1d028a55fd563a3e20561d716","internal_anchors":10},"formal_canon":{"evidence_count":2,"snapshot_sha256":"45898dc983777a68eb914b3df3c7add05f2ade743dfddc4c91658651bc751620"},"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":"2305.19118","created_at":"2026-05-18T03:30:52.354284+00:00"},{"alias_kind":"arxiv_version","alias_value":"2305.19118v4","created_at":"2026-05-18T03:30:52.354284+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2305.19118","created_at":"2026-05-18T03:30:52.354284+00:00"},{"alias_kind":"pith_short_12","alias_value":"656DEUNACXYL","created_at":"2026-05-18T12:33:33.725879+00:00"},{"alias_kind":"pith_short_16","alias_value":"656DEUNACXYL43CY","created_at":"2026-05-18T12:33:33.725879+00:00"},{"alias_kind":"pith_short_8","alias_value":"656DEUNA","created_at":"2026-05-18T12:33:33.725879+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":47,"internal_anchor_count":47,"sample":[{"citing_arxiv_id":"2506.03530","citing_title":"How Far Are We from Generating Missing Modalities with Foundation Models?","ref_index":46,"is_internal_anchor":true},{"citing_arxiv_id":"2411.15594","citing_title":"A Survey on LLM-as-a-Judge","ref_index":86,"is_internal_anchor":true},{"citing_arxiv_id":"2503.21460","citing_title":"Large Language Model Agent: A Survey on Methodology, Applications and Challenges","ref_index":80,"is_internal_anchor":true},{"citing_arxiv_id":"2504.02181","citing_title":"A Survey of Scaling in Large Language Model Reasoning","ref_index":108,"is_internal_anchor":true},{"citing_arxiv_id":"2504.12334","citing_title":"QM-ToT: A Medical Tree of Thoughts Reasoning Framework for Quantized Model","ref_index":36,"is_internal_anchor":true},{"citing_arxiv_id":"2605.22102","citing_title":"ExComm: Exploration-Stage Communication for Error-Resilient Agentic Test-Time Scaling","ref_index":28,"is_internal_anchor":true},{"citing_arxiv_id":"2310.02170","citing_title":"A Dynamic LLM-Powered Agent Network for Task-Oriented Agent Collaboration","ref_index":10,"is_internal_anchor":true},{"citing_arxiv_id":"2605.16410","citing_title":"Test-Time Hinting for Black-Box Vision-Language Models","ref_index":3,"is_internal_anchor":true},{"citing_arxiv_id":"2605.17247","citing_title":"Towards Robust Argumentative Essay Understanding via TIDE: An Interactive Framework with Trial and Debate","ref_index":80,"is_internal_anchor":true},{"citing_arxiv_id":"2605.19193","citing_title":"Sequential Consensus for Multi-Agent LLM Debates: A Wald-SPRT compute governor with calibration-based failure detection","ref_index":2,"is_internal_anchor":true},{"citing_arxiv_id":"2605.20025","citing_title":"AutoResearchClaw: Self-Reinforcing Autonomous Research with Human-AI Collaboration","ref_index":11,"is_internal_anchor":true},{"citing_arxiv_id":"2506.04565","citing_title":"From Standalone LLMs to Integrated Intelligence: A Survey of Compound Al Systems","ref_index":88,"is_internal_anchor":true},{"citing_arxiv_id":"2309.17452","citing_title":"ToRA: A Tool-Integrated Reasoning Agent for Mathematical Problem Solving","ref_index":21,"is_internal_anchor":true},{"citing_arxiv_id":"2507.00432","citing_title":"Does Math Reasoning Improve General LLM Capabilities? Understanding Transferability of LLM Reasoning","ref_index":115,"is_internal_anchor":true},{"citing_arxiv_id":"2508.15411","citing_title":"Foundational Design Principles and Patterns for Building Robust and Adaptive GenAI-Native Systems","ref_index":33,"is_internal_anchor":true},{"citing_arxiv_id":"2508.21720","citing_title":"PosterForest: Hierarchical Multi-Agent Collaboration for Scientific Poster Generation","ref_index":9,"is_internal_anchor":true},{"citing_arxiv_id":"2511.01188","citing_title":"ZoFia: Zero-Shot Fake News Detection with Entity-Guided Retrieval and Multi-LLM Interaction","ref_index":29,"is_internal_anchor":true},{"citing_arxiv_id":"2402.03578","citing_title":"LLM Multi-Agent Systems: Challenges and Open Problems","ref_index":43,"is_internal_anchor":true},{"citing_arxiv_id":"2512.04695","citing_title":"TRINITY: An Evolved LLM Coordinator","ref_index":12,"is_internal_anchor":true},{"citing_arxiv_id":"2309.02427","citing_title":"Cognitive Architectures for Language Agents","ref_index":44,"is_internal_anchor":true},{"citing_arxiv_id":"2406.04692","citing_title":"Mixture-of-Agents Enhances Large Language Model Capabilities","ref_index":15,"is_internal_anchor":true},{"citing_arxiv_id":"2309.03883","citing_title":"DoLa: Decoding by Contrasting Layers Improves Factuality in Large Language Models","ref_index":52,"is_internal_anchor":true},{"citing_arxiv_id":"2410.07283","citing_title":"Prompt Infection: LLM-to-LLM Prompt Injection within Multi-Agent Systems","ref_index":68,"is_internal_anchor":true},{"citing_arxiv_id":"2410.07283","citing_title":"Prompt Infection: LLM-to-LLM Prompt Injection within Multi-Agent Systems","ref_index":34,"is_internal_anchor":true},{"citing_arxiv_id":"2603.07880","citing_title":"What Do AI Agents Talk About? Discourse and Architectural Constraints in the First AI-Only Social Network","ref_index":29,"is_internal_anchor":true}]},"formal_canon":{"evidence_count":2,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/656DEUNACXYL43CYGYJA4Y6JXK","json":"https://pith.science/pith/656DEUNACXYL43CYGYJA4Y6JXK.json","graph_json":"https://pith.science/api/pith-number/656DEUNACXYL43CYGYJA4Y6JXK/graph.json","events_json":"https://pith.science/api/pith-number/656DEUNACXYL43CYGYJA4Y6JXK/events.json","paper":"https://pith.science/paper/656DEUNA"},"agent_actions":{"view_html":"https://pith.science/pith/656DEUNACXYL43CYGYJA4Y6JXK","download_json":"https://pith.science/pith/656DEUNACXYL43CYGYJA4Y6JXK.json","view_paper":"https://pith.science/paper/656DEUNA","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2305.19118&json=true","fetch_graph":"https://pith.science/api/pith-number/656DEUNACXYL43CYGYJA4Y6JXK/graph.json","fetch_events":"https://pith.science/api/pith-number/656DEUNACXYL43CYGYJA4Y6JXK/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/656DEUNACXYL43CYGYJA4Y6JXK/action/timestamp_anchor","attest_storage":"https://pith.science/pith/656DEUNACXYL43CYGYJA4Y6JXK/action/storage_attestation","attest_author":"https://pith.science/pith/656DEUNACXYL43CYGYJA4Y6JXK/action/author_attestation","sign_citation":"https://pith.science/pith/656DEUNACXYL43CYGYJA4Y6JXK/action/citation_signature","submit_replication":"https://pith.science/pith/656DEUNACXYL43CYGYJA4Y6JXK/action/replication_record"}},"created_at":"2026-05-18T03:30:52.354284+00:00","updated_at":"2026-05-18T03:30:52.354284+00:00"}