{"id":"https://openalex.org/W4416035861","doi":"https://doi.org/10.18653/v1/2025.findings-emnlp.112","title":"The Hallucination Tax of Reinforcement Finetuning","display_name":"The Hallucination Tax of Reinforcement Finetuning","publication_year":2025,"publication_date":"2025-01-01","ids":{"openalex":"https://openalex.org/W4416035861","doi":"https://doi.org/10.18653/v1/2025.findings-emnlp.112"},"language":null,"primary_location":{"id":"doi:10.18653/v1/2025.findings-emnlp.112","is_oa":true,"landing_page_url":"https://doi.org/10.18653/v1/2025.findings-emnlp.112","pdf_url":"https://aclanthology.org/2025.findings-emnlp.112.pdf","source":null,"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Findings of the Association for Computational Linguistics: EMNLP 2025","raw_type":"proceedings-article"},"type":"conference-paper","indexed_in":["crossref"],"open_access":{"is_oa":true,"oa_status":"gold","oa_url":"https://aclanthology.org/2025.findings-emnlp.112.pdf","any_repository_has_fulltext":null},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5100701088","display_name":"Li Song","orcid":"https://orcid.org/0000-0003-2777-7576"},"institutions":[{"id":"https://openalex.org/I1174212","display_name":"University of Southern California","ror":"https://ror.org/03taz7m60","country_code":"US","type":"education","lineage":["https://openalex.org/I1174212"]},{"id":"https://openalex.org/I2800817003","display_name":"California Southern University","ror":"https://ror.org/058zz0t50","country_code":"US","type":"education","lineage":["https://openalex.org/I2800817003"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Linxin Song","raw_affiliation_strings":["University of Southern California"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"University of Southern California","institution_ids":["https://openalex.org/I1174212","https://openalex.org/I2800817003"]}]},{"author_position":"middle","author":{"id":null,"display_name":"Taiwei Shi","orcid":null},"institutions":[{"id":"https://openalex.org/I1174212","display_name":"University of Southern California","ror":"https://ror.org/03taz7m60","country_code":"US","type":"education","lineage":["https://openalex.org/I1174212"]},{"id":"https://openalex.org/I2800817003","display_name":"California Southern University","ror":"https://ror.org/058zz0t50","country_code":"US","type":"education","lineage":["https://openalex.org/I2800817003"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Taiwei Shi","raw_affiliation_strings":["University of Southern California"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"University of Southern California","institution_ids":["https://openalex.org/I1174212","https://openalex.org/I2800817003"]}]},{"author_position":"last","author":{"id":"https://openalex.org/A5105464125","display_name":"Jieyu Zhao","orcid":null},"institutions":[{"id":"https://openalex.org/I1174212","display_name":"University of Southern California","ror":"https://ror.org/03taz7m60","country_code":"US","type":"education","lineage":["https://openalex.org/I1174212"]},{"id":"https://openalex.org/I2800817003","display_name":"California Southern University","ror":"https://ror.org/058zz0t50","country_code":"US","type":"education","lineage":["https://openalex.org/I2800817003"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Jieyu Zhao","raw_affiliation_strings":["University of Southern California"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"University of Southern California","institution_ids":["https://openalex.org/I1174212","https://openalex.org/I2800817003"]}]}],"institutions":[],"countries_distinct_count":1,"institutions_distinct_count":2,"corresponding_author_ids":[],"corresponding_institution_ids":[],"apc_list":null,"apc_paid":null,"fwci":8.037,"has_fulltext":true,"cited_by_count":1,"citation_normalized_percentile":{"value":0.96918851,"is_in_top_1_percent":false,"is_in_top_10_percent":true},"cited_by_percentile_year":{"min":94,"max":97},"biblio":{"volume":null,"issue":null,"first_page":"2105","last_page":"2120"},"is_retracted":false,"is_paratext":false,"is_xpac":false,"primary_topic":{"id":"https://openalex.org/T11762","display_name":"Law, Economics, and Judicial Systems","score":0.07750000059604645,"subfield":{"id":"https://openalex.org/subfields/2002","display_name":"Economics and Econometrics"},"field":{"id":"https://openalex.org/fields/20","display_name":"Economics, Econometrics and Finance"},"domain":{"id":"https://openalex.org/domains/2","display_name":"Social Sciences"}},"topics":[{"id":"https://openalex.org/T11762","display_name":"Law, Economics, and Judicial Systems","score":0.07750000059604645,"subfield":{"id":"https://openalex.org/subfields/2002","display_name":"Economics and Econometrics"},"field":{"id":"https://openalex.org/fields/20","display_name":"Economics, Econometrics and Finance"},"domain":{"id":"https://openalex.org/domains/2","display_name":"Social Sciences"}},{"id":"https://openalex.org/T13138","display_name":"Legal and Constitutional Studies","score":0.055399999022483826,"subfield":{"id":"https://openalex.org/subfields/2002","display_name":"Economics and Econometrics"},"field":{"id":"https://openalex.org/fields/20","display_name":"Economics, Econometrics and Finance"},"domain":{"id":"https://openalex.org/domains/2","display_name":"Social Sciences"}},{"id":"https://openalex.org/T11997","display_name":"Free Will and Agency","score":0.03060000017285347,"subfield":{"id":"https://openalex.org/subfields/2805","display_name":"Cognitive Neuroscience"},"field":{"id":"https://openalex.org/fields/28","display_name":"Neuroscience"},"domain":{"id":"https://openalex.org/domains/1","display_name":"Life Sciences"}}],"keywords":[{"id":"https://openalex.org/keywords/hallucinating","display_name":"Hallucinating","score":0.8281000256538391},{"id":"https://openalex.org/keywords/leverage","display_name":"Leverage (statistics)","score":0.722100019454956},{"id":"https://openalex.org/keywords/generalization","display_name":"Generalization","score":0.550000011920929},{"id":"https://openalex.org/keywords/electre","display_name":"ELECTRE","score":0.5346999764442444},{"id":"https://openalex.org/keywords/trustworthiness","display_name":"Trustworthiness","score":0.4672999978065491}],"concepts":[{"id":"https://openalex.org/C2911011789","wikidata":"https://www.wikidata.org/wiki/Q130741","display_name":"Hallucinating","level":2,"score":0.8281000256538391},{"id":"https://openalex.org/C153083717","wikidata":"https://www.wikidata.org/wiki/Q6535263","display_name":"Leverage (statistics)","level":2,"score":0.722100019454956},{"id":"https://openalex.org/C177148314","wikidata":"https://www.wikidata.org/wiki/Q170084","display_name":"Generalization","level":2,"score":0.550000011920929},{"id":"https://openalex.org/C47431972","wikidata":"https://www.wikidata.org/wiki/Q3045948","display_name":"ELECTRE","level":3,"score":0.5346999764442444},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.484499990940094},{"id":"https://openalex.org/C153701036","wikidata":"https://www.wikidata.org/wiki/Q659974","display_name":"Trustworthiness","level":2,"score":0.4672999978065491},{"id":"https://openalex.org/C180747234","wikidata":"https://www.wikidata.org/wiki/Q23373","display_name":"Cognitive psychology","level":1,"score":0.46560001373291016},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.4611999988555908},{"id":"https://openalex.org/C15744967","wikidata":"https://www.wikidata.org/wiki/Q9418","display_name":"Psychology","level":0,"score":0.44209998846054077},{"id":"https://openalex.org/C97541855","wikidata":"https://www.wikidata.org/wiki/Q830687","display_name":"Reinforcement learning","level":2,"score":0.43160000443458557},{"id":"https://openalex.org/C67203356","wikidata":"https://www.wikidata.org/wiki/Q1321905","display_name":"Reinforcement","level":2,"score":0.3617999851703644},{"id":"https://openalex.org/C77805123","wikidata":"https://www.wikidata.org/wiki/Q161272","display_name":"Social psychology","level":1,"score":0.3343999981880188},{"id":"https://openalex.org/C119857082","wikidata":"https://www.wikidata.org/wiki/Q2539","display_name":"Machine learning","level":1,"score":0.31450000405311584}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.18653/v1/2025.findings-emnlp.112","is_oa":true,"landing_page_url":"https://doi.org/10.18653/v1/2025.findings-emnlp.112","pdf_url":"https://aclanthology.org/2025.findings-emnlp.112.pdf","source":null,"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Findings of the Association for Computational Linguistics: EMNLP 2025","raw_type":"proceedings-article"}],"best_oa_location":{"id":"doi:10.18653/v1/2025.findings-emnlp.112","is_oa":true,"landing_page_url":"https://doi.org/10.18653/v1/2025.findings-emnlp.112","pdf_url":"https://aclanthology.org/2025.findings-emnlp.112.pdf","source":null,"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Findings of the Association for Computational Linguistics: EMNLP 2025","raw_type":"proceedings-article"},"sustainable_development_goals":[],"awards":[],"funders":[],"has_content":{"grobid_xml":true,"pdf":true},"content_urls":{"pdf":"https://content.openalex.org/works/W4416035861.pdf","grobid_xml":"https://content.openalex.org/works/W4416035861.grobid-xml"},"referenced_works_count":0,"referenced_works":[],"related_works":[],"abstract_inverted_index":{"Reinforcement":[0],"finetuning":[1],"(RFT)":[2],"has":[3],"become":[4],"a":[5,32,44,67],"standard":[6,95],"approach":[7,136],"for":[8],"enhancing":[9],"the":[10,41,87],"reasoning":[11,85],"capabilities":[12],"of":[13,36,70],"large":[14],"language":[15],"models":[16,50],"(LLMs).However,":[17],"its":[18],"impact":[19],"on":[20,132],"model":[21,100],"trustworthiness":[22],"remains":[23],"underexplored.In":[24],"this":[25,135],"work,":[26],"we":[27,39,61],"identify":[28],"and":[29,149],"systematically":[30],"study":[31],"critical":[33],"side":[34],"effect":[35],"RFT,":[37],"which":[38,107],"term":[40],"hallucination":[42],"tax:":[43],"degradation":[45],"in":[46],"refusal":[47,101,126],"behavior":[48],"causing":[49],"to":[51,55,75,79,112,139,143,156,162],"produce":[52],"hallucinated":[53],"answers":[54],"unanswerable":[56,71,82],"questions":[57],"confidently.To":[58],"investigate":[59],"this,":[60],"introduce":[62],"SUM":[63,120],"(Synthetic":[64],"Unanswerable":[65],"Math),":[66],"high-quality":[68],"dataset":[69],"math":[72,158],"problems":[73,159],"designed":[74],"probe":[76],"models'":[77],"ability":[78],"recognize":[80],"an":[81],"question":[83,164],"by":[84,103],"from":[86],"insufficient":[88],"or":[89],"ambiguous":[90],"information.Our":[91],"results":[92],"show":[93],"that":[94,116],"RFT":[96,122],"training":[97],"could":[98],"reduce":[99],"rates":[102],"more":[104],"than":[105],"80%,":[106],"significantly":[108],"increases":[109],"model's":[110],"tendency":[111],"hallucinate.We":[113],"further":[114],"demonstrate":[115],"incorporating":[117],"just":[118],"10%":[119],"during":[121],"substantially":[123],"restores":[124],"appropriate":[125],"behavior,":[127],"with":[128],"minimal":[129],"accuracy":[130],"trade-offs":[131],"solvable":[133],"tasks.Crucially,":[134],"enables":[137],"LLMs":[138],"leverage":[140],"inference-time":[141],"compute":[142],"reason":[144],"about":[145],"their":[146],"own":[147],"uncertainty":[148],"knowledge":[150],"boundaries,":[151],"improving":[152],"generalization":[153],"not":[154],"only":[155],"outof-domain":[157],"but":[160],"also":[161],"factual":[163],"answering":[165],"tasks.":[166]},"counts_by_year":[{"year":2026,"cited_by_count":1}],"updated_date":"2026-08-21T09:56:20.448147","created_date":"2025-11-08T00:00:00"}