{"id":"https://openalex.org/W7148546791","doi":"https://doi.org/10.48550/arxiv.2604.00304","title":"Asymmetric Actor-Critic for Multi-turn LLM Agents","display_name":"Asymmetric Actor-Critic for Multi-turn LLM Agents","publication_year":2026,"publication_date":"2026-03-31","ids":{"openalex":"https://openalex.org/W7148546791","doi":"https://doi.org/10.48550/arxiv.2604.00304"},"language":null,"primary_location":{"id":"doi:10.48550/arxiv.2604.00304","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2604.00304","pdf_url":null,"source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":null,"issn":null,"is_oa":true,"is_in_doaj":false,"is_core":false,"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":null,"license_id":null,"version":null,"is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"article"},"type":"preprint","indexed_in":["datacite"],"open_access":{"is_oa":true,"oa_status":"green","oa_url":"https://doi.org/10.48550/arxiv.2604.00304","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5127477129","display_name":"Shuli Jiang","orcid":null},"institutions":[],"countries":[],"is_corresponding":true,"raw_author_name":"Jiang, Shuli","raw_affiliation_strings":[],"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5132817883","display_name":"Zhaoyang Zhang","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Zhang, Zhaoyang","raw_affiliation_strings":[],"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5132827376","display_name":"Yi Zhang","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Zhang, Yi","raw_affiliation_strings":[],"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5132828791","display_name":"shuo yang","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Yang, Shuo","raw_affiliation_strings":[],"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5132793245","display_name":"Wei Xia","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Xia, Wei","raw_affiliation_strings":[],"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5132819831","display_name":"Stefano Soatto","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Soatto, Stefano","raw_affiliation_strings":[],"affiliations":[]}],"institutions":[],"countries_distinct_count":0,"institutions_distinct_count":6,"corresponding_author_ids":["https://openalex.org/A5127477129"],"corresponding_institution_ids":[],"apc_list":null,"apc_paid":null,"fwci":null,"has_fulltext":false,"cited_by_count":0,"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":0.6592000126838684,"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":0.6592000126838684,"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/T11714","display_name":"Multimodal Machine Learning Applications","score":0.0877000018954277,"subfield":{"id":"https://openalex.org/subfields/1707","display_name":"Computer Vision and Pattern Recognition"},"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/T11636","display_name":"Artificial Intelligence in Healthcare and Education","score":0.0414000004529953,"subfield":{"id":"https://openalex.org/subfields/2718","display_name":"Health Informatics"},"field":{"id":"https://openalex.org/fields/27","display_name":"Medicine"},"domain":{"id":"https://openalex.org/domains/4","display_name":"Health Sciences"}}],"keywords":[{"id":"https://openalex.org/keywords/leverage","display_name":"Leverage (statistics)","score":0.7731000185012817},{"id":"https://openalex.org/keywords/task","display_name":"Task (project management)","score":0.5113999843597412},{"id":"https://openalex.org/keywords/reliability","display_name":"Reliability (semiconductor)","score":0.5016000270843506},{"id":"https://openalex.org/keywords/reflection","display_name":"Reflection (computer programming)","score":0.49959999322891235},{"id":"https://openalex.org/keywords/pipeline","display_name":"Pipeline (software)","score":0.4846999943256378},{"id":"https://openalex.org/keywords/inference","display_name":"Inference","score":0.34689998626708984}],"concepts":[{"id":"https://openalex.org/C153083717","wikidata":"https://www.wikidata.org/wiki/Q6535263","display_name":"Leverage (statistics)","level":2,"score":0.7731000185012817},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.7340999841690063},{"id":"https://openalex.org/C2780451532","wikidata":"https://www.wikidata.org/wiki/Q759676","display_name":"Task (project management)","level":2,"score":0.5113999843597412},{"id":"https://openalex.org/C43214815","wikidata":"https://www.wikidata.org/wiki/Q7310987","display_name":"Reliability (semiconductor)","level":3,"score":0.5016000270843506},{"id":"https://openalex.org/C65682993","wikidata":"https://www.wikidata.org/wiki/Q1056451","display_name":"Reflection (computer programming)","level":2,"score":0.49959999322891235},{"id":"https://openalex.org/C43521106","wikidata":"https://www.wikidata.org/wiki/Q2165493","display_name":"Pipeline (software)","level":2,"score":0.4846999943256378},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.4153999984264374},{"id":"https://openalex.org/C107457646","wikidata":"https://www.wikidata.org/wiki/Q207434","display_name":"Human\u2013computer interaction","level":1,"score":0.35359999537467957},{"id":"https://openalex.org/C2776214188","wikidata":"https://www.wikidata.org/wiki/Q408386","display_name":"Inference","level":2,"score":0.34689998626708984},{"id":"https://openalex.org/C175154964","wikidata":"https://www.wikidata.org/wiki/Q380077","display_name":"Task analysis","level":3,"score":0.2865999937057495},{"id":"https://openalex.org/C137293760","wikidata":"https://www.wikidata.org/wiki/Q3621696","display_name":"Language model","level":2,"score":0.28360000252723694},{"id":"https://openalex.org/C2779343474","wikidata":"https://www.wikidata.org/wiki/Q3109175","display_name":"Context (archaeology)","level":2,"score":0.2696000039577484},{"id":"https://openalex.org/C119857082","wikidata":"https://www.wikidata.org/wiki/Q2539","display_name":"Machine learning","level":1,"score":0.26570001244544983}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.48550/arxiv.2604.00304","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2604.00304","pdf_url":null,"source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":null,"issn":null,"is_oa":true,"is_in_doaj":false,"is_core":false,"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":null,"license_id":null,"version":null,"is_accepted":false,"is_published":null,"raw_source_name":null,"raw_type":"article"}],"best_oa_location":{"id":"doi:10.48550/arxiv.2604.00304","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2604.00304","pdf_url":null,"source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":null,"issn":null,"is_oa":true,"is_in_doaj":false,"is_core":false,"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":null,"license_id":null,"version":null,"is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"article"},"sustainable_development_goals":[{"score":0.5506076812744141,"display_name":"Quality Education","id":"https://metadata.un.org/sdg/4"}],"awards":[],"funders":[],"has_content":{"pdf":false,"grobid_xml":false},"content_urls":null,"referenced_works_count":0,"referenced_works":[],"related_works":[],"abstract_inverted_index":{"Large":[0],"language":[1],"models":[2,50,175],"(LLMs)":[3],"exhibit":[4],"strong":[5,163],"reasoning":[6],"and":[7,86,150,159,180],"conversational":[8,64,106],"abilities,":[9],"but":[10],"ensuring":[11],"reliable":[12,63],"behavior":[13],"in":[14,26,104,176],"multi-turn":[15],"interactions":[16],"remains":[17],"challenging.":[18],"In":[19],"many":[20],"real-world":[21],"applications,":[22],"agents":[23],"must":[24],"succeed":[25],"one-shot":[27],"settings":[28],"where":[29],"retries":[30],"are":[31],"impossible.":[32],"Existing":[33],"approaches":[34],"either":[35],"rely":[36],"on":[37,148],"reflection":[38],"or":[39,46,171],"post-hoc":[40],"evaluation,":[41],"which":[42],"require":[43],"additional":[44,184],"attempts,":[45],"assume":[47],"fully":[48],"trainable":[49],"that":[51,136,153],"cannot":[52],"leverage":[53],"proprietary":[54,68,174],"LLMs.":[55],"We":[56,129],"propose":[57],"an":[58],"asymmetric":[59],"actor-critic":[60,95],"framework":[61,98],"for":[62,140],"agents.":[65],"A":[66],"powerful":[67],"LLM":[69],"acts":[70],"as":[71],"the":[72,83,89,145,177],"actor,":[73],"while":[74,114],"a":[75,100,111,132],"smaller":[76,127],"open-source":[77,168],"critic":[78,141,178,181],"provides":[79],"runtime":[80],"supervision,":[81],"monitoring":[82],"actor's":[84],"actions":[85],"intervening":[87],"within":[88],"same":[90],"interaction":[91],"trajectory.":[92],"Unlike":[93],"training-based":[94],"methods,":[96],"our":[97,154],"supervises":[99],"fixed":[101],"actor":[102],"operating":[103],"open-ended":[105],"environments.":[107],"The":[108],"design":[109],"leverages":[110],"generation-verification":[112],"asymmetry:":[113],"high-quality":[115],"generation":[116,134],"requires":[117],"large":[118],"models,":[119],"effective":[120],"oversight":[121],"can":[122],"often":[123],"be":[124],"achieved":[125],"by":[126],"ones.":[128],"further":[130],"introduce":[131],"data":[133],"pipeline":[135],"produces":[137],"supervision":[138],"signals":[139],"fine-tuning":[142,182],"without":[143],"modifying":[144],"actor.":[146],"Experiments":[147],"$\u03c4$-bench":[149],"UserBench":[151],"show":[152],"approach":[155],"significantly":[156],"improves":[157],"reliability":[158],"task":[160],"success":[161],"over":[162,186],"single-agent":[164],"baselines.":[165],"Moreover,":[166],"lightweight":[167],"critics":[169],"rival":[170],"surpass":[172],"larger":[173],"role,":[179],"yields":[183],"gains":[185],"several":[187],"state-of-the-art":[188],"methods.":[189]},"counts_by_year":[],"updated_date":"2026-04-03T16:44:17.987007","created_date":"2026-04-03T00:00:00"}