{"id":"https://openalex.org/W4389217455","doi":"https://doi.org/10.48550/arxiv.2311.17371","title":"Should we be going MAD? A Look at Multi-Agent Debate Strategies for LLMs","display_name":"Should we be going MAD? A Look at Multi-Agent Debate Strategies for LLMs","publication_year":2023,"publication_date":"2023-11-29","ids":{"openalex":"https://openalex.org/W4389217455","doi":"https://doi.org/10.48550/arxiv.2311.17371"},"language":"en","primary_location":{"id":"pmh:oai:arXiv.org:2311.17371","doi":null,"is_oa":true,"landing_page_url":"http://arxiv.org/abs/2311.17371","pdf_url":"https://arxiv.org/pdf/2311.17371","source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":"2331-8422","issn":["2331-8422"],"is_oa":true,"is_in_doaj":false,"is_core":false,"listed_in":[],"host_organization":"https://openalex.org/I205783295","host_organization_name":"Cornell University","host_organization_lineage":["https://openalex.org/I205783295"],"host_organization_lineage_names":[],"type":"repository"},"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"text"},"volume":null,"type":"preprint","indexed_in":["arxiv","datacite"],"open_access":{"is_oa":true,"oa_status":"green","oa_url":"https://arxiv.org/pdf/2311.17371","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5046370375","display_name":"Andries Smit","orcid":"https://orcid.org/0000-0003-2796-5497","observed_orcids":["https://orcid.org/0000-0003-2796-5497"]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Smit, Andries","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5063524295","display_name":"Paul Duckworth","orcid":"https://orcid.org/0000-0001-9052-6919","observed_orcids":["https://orcid.org/0000-0001-9052-6919"]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Duckworth, Paul","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5032092481","display_name":"Nathan Grinsztajn","orcid":"https://orcid.org/0000-0001-6817-5972","observed_orcids":["https://orcid.org/0000-0001-6817-5972"]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Grinsztajn, Nathan","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5052796575","display_name":"Thomas D. Barrett","orcid":"https://orcid.org/0000-0001-6241-3028","observed_orcids":["https://orcid.org/0000-0001-6241-3028"]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Barrett, Thomas D.","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5017043594","display_name":"Arnu Pretorius","orcid":null,"observed_orcids":[]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Pretorius, Arnu","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]}],"institutions":[],"countries_distinct_count":0,"institutions_distinct_count":0,"corresponding_author_ids":[],"corresponding_institution_ids":[],"apc_list":null,"apc_paid":null,"fwci":null,"has_fulltext":true,"cited_by_count":6,"citation_normalized_percentile":null,"cited_by_percentile_year":null,"biblio":{"volume":null,"issue":null,"first_page":null,"last_page":null},"is_retracted":false,"is_paratext":false,"is_xpac":false,"primary_topic":{"id":"https://openalex.org/T10028","display_name":"Topic Modeling","score":0.5580999851226807,"subfield":{"id":"https://openalex.org/subfields/1702","display_name":"Artificial Intelligence"},"field":{"id":"https://openalex.org/fields/17","display_name":"Computer Science"},"domain":{"id":"https://openalex.org/domains/3","display_name":"Physical Sciences"}},"topics":[{"id":"https://openalex.org/T10028","display_name":"Topic Modeling","score":0.5580999851226807,"subfield":{"id":"https://openalex.org/subfields/1702","display_name":"Artificial Intelligence"},"field":{"id":"https://openalex.org/fields/17","display_name":"Computer Science"},"domain":{"id":"https://openalex.org/domains/3","display_name":"Physical Sciences"}},{"id":"https://openalex.org/T10181","display_name":"Natural Language Processing Techniques","score":0.09910000115633011,"subfield":{"id":"https://openalex.org/subfields/1702","display_name":"Artificial Intelligence"},"field":{"id":"https://openalex.org/fields/17","display_name":"Computer Science"},"domain":{"id":"https://openalex.org/domains/3","display_name":"Physical Sciences"}},{"id":"https://openalex.org/T10456","display_name":"Multi-Agent Systems and Negotiation","score":0.06880000233650208,"subfield":{"id":"https://openalex.org/subfields/1702","display_name":"Artificial Intelligence"},"field":{"id":"https://openalex.org/fields/17","display_name":"Computer Science"},"domain":{"id":"https://openalex.org/domains/3","display_name":"Physical Sciences"}}],"keywords":[{"id":"https://openalex.org/keywords/multi-agent-debate","display_name":"multi-agent debate","score":0.9879999756813049},{"id":"https://openalex.org/keywords/large-language-models","display_name":"large language models","score":0.9480000138282776},{"id":"https://openalex.org/keywords/honesty","display_name":"honesty","score":0.7770000100135803},{"id":"https://openalex.org/keywords/prompting-strategies","display_name":"prompting strategies","score":0.6150000095367432},{"id":"https://openalex.org/keywords/self-consistency","display_name":"self-consistency","score":0.9049999713897705},{"id":"https://openalex.org/keywords/ensembling","display_name":"ensembling","score":0.6349999904632568},{"id":"https://openalex.org/keywords/hyperparameter-optimization","display_name":"hyperparameter optimization","score":0.8050000071525574},{"id":"https://openalex.org/keywords/accuracy-cost-trade-off","display_name":"accuracy-cost trade-off","score":0.6629999876022339}],"concepts":[{"id":"https://openalex.org/C86251818","wikidata":"https://www.wikidata.org/wiki/Q816754","display_name":"Benchmarking","level":2,"score":0.8226683735847473},{"id":"https://openalex.org/C185798385","wikidata":"https://www.wikidata.org/wiki/Q1161707","display_name":"Benchmark (surveying)","level":2,"score":0.7464938163757324},{"id":"https://openalex.org/C8642999","wikidata":"https://www.wikidata.org/wiki/Q4171168","display_name":"Hyperparameter","level":2,"score":0.7083168029785156},{"id":"https://openalex.org/C2776436953","wikidata":"https://www.wikidata.org/wiki/Q5163215","display_name":"Consistency (knowledge bases)","level":2,"score":0.6907915472984314},{"id":"https://openalex.org/C2779343474","wikidata":"https://www.wikidata.org/wiki/Q3109175","display_name":"Context (archaeology)","level":2,"score":0.6705695390701294},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.6565275192260742},{"id":"https://openalex.org/C61423126","wikidata":"https://www.wikidata.org/wiki/Q187432","display_name":"Scripting language","level":2,"score":0.5396071076393127},{"id":"https://openalex.org/C39890363","wikidata":"https://www.wikidata.org/wiki/Q36108","display_name":"Generative grammar","level":2,"score":0.4756176769733429},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.4675648808479309},{"id":"https://openalex.org/C119857082","wikidata":"https://www.wikidata.org/wiki/Q2539","display_name":"Machine learning","level":1,"score":0.4439792037010193},{"id":"https://openalex.org/C2522767166","wikidata":"https://www.wikidata.org/wiki/Q2374463","display_name":"Data science","level":1,"score":0.3934161961078644},{"id":"https://openalex.org/C144133560","wikidata":"https://www.wikidata.org/wiki/Q4830453","display_name":"Business","level":0,"score":0.13255950808525085},{"id":"https://openalex.org/C162853370","wikidata":"https://www.wikidata.org/wiki/Q39809","display_name":"Marketing","level":1,"score":0.08792215585708618},{"id":"https://openalex.org/C111919701","wikidata":"https://www.wikidata.org/wiki/Q9135","display_name":"Operating system","level":1,"score":0.0},{"id":"https://openalex.org/C86803240","wikidata":"https://www.wikidata.org/wiki/Q420","display_name":"Biology","level":0,"score":0.0},{"id":"https://openalex.org/C151730666","wikidata":"https://www.wikidata.org/wiki/Q7205","display_name":"Paleontology","level":1,"score":0.0},{"id":"https://openalex.org/C205649164","wikidata":"https://www.wikidata.org/wiki/Q1071","display_name":"Geography","level":0,"score":0.0},{"id":"https://openalex.org/C13280743","wikidata":"https://www.wikidata.org/wiki/Q131089","display_name":"Geodesy","level":1,"score":0.0}],"mesh":[],"locations_count":2,"locations":[{"id":"pmh:oai:arXiv.org:2311.17371","doi":null,"is_oa":true,"landing_page_url":"http://arxiv.org/abs/2311.17371","pdf_url":"https://arxiv.org/pdf/2311.17371","source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":"2331-8422","issn":["2331-8422"],"is_oa":true,"is_in_doaj":false,"is_core":false,"listed_in":[],"host_organization":"https://openalex.org/I205783295","host_organization_name":"Cornell University","host_organization_lineage":["https://openalex.org/I205783295"],"host_organization_lineage_names":[],"type":"repository"},"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"text"},{"id":"doi:10.48550/arxiv.2311.17371","doi":"https://doi.org/10.48550/arxiv.2311.17371","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2311.17371","pdf_url":null,"source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":"2331-8422","issn":["2331-8422"],"is_oa":true,"is_in_doaj":false,"is_core":false,"listed_in":[],"host_organization":"https://openalex.org/I205783295","host_organization_name":"Cornell University","host_organization_lineage":["https://openalex.org/I205783295"],"host_organization_lineage_names":[],"type":"repository"},"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":null,"is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"Preprint"}],"best_oa_location":{"id":"pmh:oai:arXiv.org:2311.17371","doi":null,"is_oa":true,"landing_page_url":"http://arxiv.org/abs/2311.17371","pdf_url":"https://arxiv.org/pdf/2311.17371","source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":"2331-8422","issn":["2331-8422"],"is_oa":true,"is_in_doaj":false,"is_core":false,"listed_in":[],"host_organization":"https://openalex.org/I205783295","host_organization_name":"Cornell University","host_organization_lineage":["https://openalex.org/I205783295"],"host_organization_lineage_names":[],"type":"repository"},"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"text"},"sustainable_development_goals":[],"study_designs":[],"awards":[],"funders":[],"has_content":{"grobid_xml":true,"pdf":true},"content_urls":{"pdf":"https://content.openalex.org/works/W4389217455.pdf","grobid_xml":"https://content.openalex.org/works/W4389217455.grobid-xml"},"referenced_works_count":0,"referenced_works":[],"related_works":["https://openalex.org/W7166837933","https://openalex.org/W7101394647","https://openalex.org/W4416260560","https://openalex.org/W7215091782","https://openalex.org/W4412888437","https://openalex.org/W7171878187","https://openalex.org/W4416344031","https://openalex.org/W4416189552","https://openalex.org/W4415087697","https://openalex.org/W7155082204"],"abstract_inverted_index":{"Recent":[0],"advancements":[1],"in":[2,14,74],"large":[3],"language":[4],"models":[5],"(LLMs)":[6],"underscore":[7],"their":[8,75],"potential":[9],"for":[10,43],"responding":[11],"to":[12,58,127,133,140,172,183],"inquiries":[13],"various":[15],"domains.":[16],"However,":[17,95],"ensuring":[18],"that":[19,70,110,122],"generative":[20],"agents":[21],"provide":[22,168],"accurate":[23],"and":[24,55,65,89,131,158],"reliable":[25],"answers":[26],"remains":[27],"an":[28,169],"ongoing":[29],"challenge.":[30],"In":[31],"this":[32],"context,":[33],"multi-agent":[34,71],"debate":[35],"(MAD)":[36],"has":[37],"emerged":[38],"as":[39,87,104,148],"a":[40,51],"promising":[41],"strategy":[42],"enhancing":[44],"the":[45,60,173],"truthfulness":[46],"of":[47,53],"LLMs.":[48],"We":[49,135,167],"benchmark":[50,184],"range":[52],"debating":[54,72,145],"prompting":[56,84],"strategies":[57],"explore":[59],"trade-offs":[61],"between":[62],"cost,":[63],"time,":[64],"accuracy.":[66],"Importantly,":[67],"we":[68,165],"find":[69],"systems,":[73,102],"current":[76],"form,":[77],"do":[78],"not":[79,114],"reliably":[80],"outperform":[81],"other":[82,119,162],"proposed":[83],"strategies,":[85,146],"such":[86,103,147],"self-consistency":[88],"ensembling":[90],"using":[91],"multiple":[92],"reasoning":[93],"paths.":[94],"when":[96],"performing":[97],"hyperparameter":[98,129],"tuning,":[99],"several":[100,176],"MAD":[101,111],"Multi-Persona,":[105],"perform":[106],"better.":[107],"This":[108],"suggests":[109],"protocols":[112,164,178],"might":[113],"be":[115],"inherently":[116],"worse":[117],"than":[118],"approaches,":[120],"but":[121],"they":[123],"are":[124],"more":[125],"sensitive":[126],"different":[128],"settings":[130],"difficult":[132],"optimize.":[134],"build":[136],"on":[137],"these":[138],"results":[139],"offer":[141],"insights":[142],"into":[143],"improving":[144],"adjusting":[149],"agent":[150],"agreement":[151],"levels,":[152],"which":[153],"can":[154],"significantly":[155],"enhance":[156],"performance":[157],"even":[159],"surpass":[160],"all":[161],"non-debate":[163],"evaluated.":[166],"open-source":[170],"repository":[171],"community":[174],"with":[175,180],"state-of-the-art":[177],"together":[179],"evaluation":[181],"scripts":[182],"across":[185],"popular":[186],"research":[187],"datasets.":[188]},"counts_by_year":[{"year":2026,"cited_by_count":4},{"year":2025,"cited_by_count":2}],"updated_date":"2026-10-09T09:14:12.587421","created_date":"2023-12-01T00:00:00"}