{"id":"https://openalex.org/W7214746446","doi":"https://doi.org/10.48550/arxiv.2609.32470","title":"On the Pitfalls of Verbalized Confidence Priors for Calibrating Large Reasoning Models","display_name":"On the Pitfalls of Verbalized Confidence Priors for Calibrating Large Reasoning Models","publication_year":2026,"publication_date":"2026-09-26","ids":{"openalex":"https://openalex.org/W7214746446","doi":"https://doi.org/10.48550/arxiv.2609.32470"},"language":"en","primary_location":{"id":"pmh:oai:arXiv.org:2609.32470","doi":null,"is_oa":true,"landing_page_url":"https://arxiv.org/abs/2609.32470","pdf_url":"https://arxiv.org/pdf/2609.32470","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":null,"license_id":null,"version":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"text"},"volume":null,"type":"preprint","is_traditional":true,"indexed_in":["arxiv","datacite"],"open_access":{"is_oa":true,"oa_status":"green","oa_url":"https://arxiv.org/pdf/2609.32470","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5153548261","display_name":"Shuoyuan Wang","orcid":null,"observed_orcids":[]},"institutions":[{"id":"https://openalex.org/I3045169105","display_name":"Southern University of Science and Technology","ror":"https://ror.org/049tv2d57","country_code":"CN","type":"education","lineage":["https://openalex.org/I3045169105"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Wang, Shuoyuan","raw_affiliation_strings":["Southern University of Science and Technology"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Southern University of Science and Technology","institution_ids":["https://openalex.org/I3045169105"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5122592059","display_name":"Beier Luo","orcid":null,"observed_orcids":[]},"institutions":[{"id":"https://openalex.org/I172675005","display_name":"Nanyang Technological University","ror":"https://ror.org/02e7b5302","country_code":"SG","type":"education","lineage":["https://openalex.org/I172675005"]}],"countries":["SG"],"is_corresponding":false,"raw_author_name":"Luo, Beier","raw_affiliation_strings":["Nanyang Technological University"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Nanyang Technological University","institution_ids":["https://openalex.org/I172675005"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5153502735","display_name":"Hao Zeng","orcid":null,"observed_orcids":[]},"institutions":[{"id":"https://openalex.org/I150229711","display_name":"University of Electronic Science and Technology of China","ror":"https://ror.org/04qr3zq92","country_code":"CN","type":"education","lineage":["https://openalex.org/I150229711"]},{"id":"https://openalex.org/I3045169105","display_name":"Southern University of Science and Technology","ror":"https://ror.org/049tv2d57","country_code":"CN","type":"education","lineage":["https://openalex.org/I3045169105"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Zeng, Hao","raw_affiliation_strings":["Southern University of Science and Technology","University of Electronic Science and Technology of China"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Southern University of Science and Technology","institution_ids":["https://openalex.org/I3045169105"]},{"raw_affiliation_string":"University of Electronic Science and Technology of China","institution_ids":["https://openalex.org/I150229711"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5124756145","display_name":"Chengyao Yu","orcid":null,"observed_orcids":[]},"institutions":[{"id":"https://openalex.org/I3045169105","display_name":"Southern University of Science and Technology","ror":"https://ror.org/049tv2d57","country_code":"CN","type":"education","lineage":["https://openalex.org/I3045169105"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Yu, Chengyao","raw_affiliation_strings":["Southern University of Science and Technology"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Southern University of Science and Technology","institution_ids":["https://openalex.org/I3045169105"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5153402945","display_name":"Songxin Zhang","orcid":null,"observed_orcids":[]},"institutions":[{"id":"https://openalex.org/I3045169105","display_name":"Southern University of Science and Technology","ror":"https://ror.org/049tv2d57","country_code":"CN","type":"education","lineage":["https://openalex.org/I3045169105"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Zhang, Songxin","raw_affiliation_strings":["Southern University of Science and Technology"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Southern University of Science and Technology","institution_ids":["https://openalex.org/I3045169105"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5114126142","display_name":"Zejian Xie","orcid":"https://orcid.org/0009-0003-2967-0526","observed_orcids":["https://orcid.org/0009-0003-2967-0526"]},"institutions":[{"id":"https://openalex.org/I3045169105","display_name":"Southern University of Science and Technology","ror":"https://ror.org/049tv2d57","country_code":"CN","type":"education","lineage":["https://openalex.org/I3045169105"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Xie, Zejian","raw_affiliation_strings":["Southern University of Science and Technology"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Southern University of Science and Technology","institution_ids":["https://openalex.org/I3045169105"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5075871750","display_name":"Bing\u2010Yi Jing","orcid":"https://orcid.org/0000-0002-8876-1570","observed_orcids":["https://orcid.org/0000-0002-8876-1570"]},"institutions":[{"id":"https://openalex.org/I4210116924","display_name":"Chinese University of Hong Kong, Shenzhen","ror":"https://ror.org/02d5ks197","country_code":"CN","type":"education","lineage":["https://openalex.org/I177725633","https://openalex.org/I180726961","https://openalex.org/I4210116924"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Jing, Bingyi","raw_affiliation_strings":["Shenzhen Loop Area Institute","The Chinese University of Hong Kong , Shenzhen"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Shenzhen Loop Area Institute","institution_ids":[]},{"raw_affiliation_string":"The Chinese University of Hong Kong , Shenzhen","institution_ids":["https://openalex.org/I4210116924"]}]},{"author_position":"last","author":{"id":"https://openalex.org/A5153487146","display_name":"Hongxin Wei","orcid":null,"observed_orcids":[]},"institutions":[{"id":"https://openalex.org/I3045169105","display_name":"Southern University of Science and Technology","ror":"https://ror.org/049tv2d57","country_code":"CN","type":"education","lineage":["https://openalex.org/I3045169105"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Wei, Hongxin","raw_affiliation_strings":["Southern University of Science and Technology"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Southern University of Science and Technology","institution_ids":["https://openalex.org/I3045169105"]}]}],"institutions":[],"countries_distinct_count":2,"institutions_distinct_count":4,"corresponding_author_ids":[],"corresponding_institution_ids":[],"apc_list":null,"apc_paid":null,"fwci":null,"has_fulltext":true,"cited_by_count":1,"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/T11689","display_name":"Adversarial Robustness in Machine Learning","score":0.4113999903202057,"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/T11689","display_name":"Adversarial Robustness in Machine Learning","score":0.4113999903202057,"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/T10028","display_name":"Topic Modeling","score":0.1995999962091446,"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/T12072","display_name":"Machine Learning and Algorithms","score":0.0714000016450882,"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/confidence-calibration","display_name":"confidence calibration","score":0.6729999780654907},{"id":"https://openalex.org/keywords/large-reasoning-models","display_name":"large reasoning models","score":0.765999972820282},{"id":"https://openalex.org/keywords/overconfidence","display_name":"overconfidence","score":0.609000027179718},{"id":"https://openalex.org/keywords/supervised-fine-tuning","display_name":"supervised fine-tuning","score":0.9340000152587891},{"id":"https://openalex.org/keywords/proper-scoring-rules","display_name":"proper scoring rules","score":0.7429999709129333},{"id":"https://openalex.org/keywords/selective-prediction","display_name":"selective prediction","score":0.9070000052452087},{"id":"https://openalex.org/keywords/model-routing","display_name":"model routing","score":0.8640000224113464},{"id":"https://openalex.org/keywords/mathematical-reasoning","display_name":"mathematical reasoning","score":0.703000009059906}],"concepts":[{"id":"https://openalex.org/C51110983","wikidata":"https://www.wikidata.org/wiki/Q16503490","display_name":"Overconfidence effect","level":2,"score":0.805899977684021},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.6880999803543091},{"id":"https://openalex.org/C2909755999","wikidata":"https://www.wikidata.org/wiki/Q4751126","display_name":"Low Confidence","level":2,"score":0.6074000000953674},{"id":"https://openalex.org/C165838908","wikidata":"https://www.wikidata.org/wiki/Q736777","display_name":"Calibration","level":2,"score":0.5569000244140625},{"id":"https://openalex.org/C13662513","wikidata":"https://www.wikidata.org/wiki/Q5160087","display_name":"Confidence distribution","level":3,"score":0.5180000066757202},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.5175999999046326},{"id":"https://openalex.org/C177769412","wikidata":"https://www.wikidata.org/wiki/Q278090","display_name":"Prior probability","level":3,"score":0.4537000060081482},{"id":"https://openalex.org/C119857082","wikidata":"https://www.wikidata.org/wiki/Q2539","display_name":"Machine learning","level":1,"score":0.45249998569488525},{"id":"https://openalex.org/C44249647","wikidata":"https://www.wikidata.org/wiki/Q208498","display_name":"Confidence interval","level":2,"score":0.40779998898506165},{"id":"https://openalex.org/C35405484","wikidata":"https://www.wikidata.org/wiki/Q4967066","display_name":"Brier score","level":2,"score":0.40549999475479126},{"id":"https://openalex.org/C97541855","wikidata":"https://www.wikidata.org/wiki/Q830687","display_name":"Reinforcement learning","level":2,"score":0.40299999713897705},{"id":"https://openalex.org/C61797465","wikidata":"https://www.wikidata.org/wiki/Q1188986","display_name":"Term (time)","level":2,"score":0.3702000081539154},{"id":"https://openalex.org/C75553542","wikidata":"https://www.wikidata.org/wiki/Q178161","display_name":"A priori and a posteriori","level":2,"score":0.3215000033378601},{"id":"https://openalex.org/C143380155","wikidata":"https://www.wikidata.org/wiki/Q5160088","display_name":"Confidence region","level":3,"score":0.3165999948978424},{"id":"https://openalex.org/C204323151","wikidata":"https://www.wikidata.org/wiki/Q905424","display_name":"Range (aeronautics)","level":2,"score":0.28049999475479126},{"id":"https://openalex.org/C51632099","wikidata":"https://www.wikidata.org/wiki/Q3985153","display_name":"Training set","level":2,"score":0.2743000090122223},{"id":"https://openalex.org/C2776292839","wikidata":"https://www.wikidata.org/wiki/Q5179217","display_name":"Coverage probability","level":3,"score":0.2694999873638153},{"id":"https://openalex.org/C2776325391","wikidata":"https://www.wikidata.org/wiki/Q6917865","display_name":"Motivated reasoning","level":3,"score":0.2506999969482422}],"mesh":[],"locations_count":2,"locations":[{"id":"pmh:oai:arXiv.org:2609.32470","doi":null,"is_oa":true,"landing_page_url":"https://arxiv.org/abs/2609.32470","pdf_url":"https://arxiv.org/pdf/2609.32470","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":null,"license_id":null,"version":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"text"},{"id":"doi:10.48550/arxiv.2609.32470","doi":"https://doi.org/10.48550/arxiv.2609.32470","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2609.32470","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:2609.32470","doi":null,"is_oa":true,"landing_page_url":"https://arxiv.org/abs/2609.32470","pdf_url":"https://arxiv.org/pdf/2609.32470","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":null,"license_id":null,"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/W7214746446.pdf","grobid_xml":"https://content.openalex.org/works/W7214746446.grobid-xml"},"referenced_works_count":42,"referenced_works":["https://openalex.org/W2025720061","https://openalex.org/W3210904070","https://openalex.org/W4285429195","https://openalex.org/W4321594373","https://openalex.org/W4389518686","https://openalex.org/W4391631327","https://openalex.org/W4402671952","https://openalex.org/W4404783656","https://openalex.org/W4405630972","https://openalex.org/W4407039591","https://openalex.org/W4410803261","https://openalex.org/W4412887774","https://openalex.org/W4414281281","https://openalex.org/W4414855266","https://openalex.org/W4414902044","https://openalex.org/W4414939355","https://openalex.org/W4416048648","https://openalex.org/W4416073589","https://openalex.org/W4416763768","https://openalex.org/W4417465430","https://openalex.org/W6910644203","https://openalex.org/W7092283956","https://openalex.org/W7125907824","https://openalex.org/W7129460596","https://openalex.org/W7134248873","https://openalex.org/W7134854408","https://openalex.org/W7134981730","https://openalex.org/W7139956085","https://openalex.org/W7140111758","https://openalex.org/W7156330406","https://openalex.org/W7161010363","https://openalex.org/W7161579489","https://openalex.org/W7161773685","https://openalex.org/W7162341483","https://openalex.org/W7164409435","https://openalex.org/W7165750871","https://openalex.org/W7166552259","https://openalex.org/W7167276854","https://openalex.org/W7167585751","https://openalex.org/W7167604932","https://openalex.org/W7196953391","https://openalex.org/W7201914019"],"related_works":["https://openalex.org/W7168786963","https://openalex.org/W4414427575","https://openalex.org/W7221153679","https://openalex.org/W7165750871","https://openalex.org/W7156853946","https://openalex.org/W4415024084","https://openalex.org/W7127371570","https://openalex.org/W7214792109","https://openalex.org/W7163426460","https://openalex.org/W6891889810"],"abstract_inverted_index":{"Large":[0],"reasoning":[1,170,180],"models":[2],"(LRMs)":[3],"often":[4],"suffer":[5],"from":[6,152],"overconfidence":[7],"when":[8],"expressing":[9],"their":[10,157],"uncertainty.":[11],"Confidence-aware":[12],"reinforcement":[13],"learning":[14],"(RL)":[15],"offers":[16],"a":[17,53,59,99,106],"promising":[18],"way":[19],"to":[20,143],"optimize":[21],"calibration.":[22],"However,":[23],"it":[24],"relies":[25],"on":[26,58,87,172],"on-policy":[27],"rollouts":[28],"and":[29,82,134,178,187,208],"is":[30,213],"thus":[31],"constrained":[32],"by":[33],"the":[34,84,140],"model's":[35],"pre-RL":[36],"confidence":[37,42,54,80,108,120,141,146,164],"distribution,":[38],"which":[39,63,129],"we":[40,47,68,96],"term":[41],"prior.":[43],"In":[44],"this":[45,71,93],"work,":[46],"reveal":[48],"that":[49,70,104],"off-the-shelf":[50],"LRMs":[51],"exhibit":[52],"prior":[55,109],"heavily":[56],"concentrated":[57],"few":[60],"high":[61],"values,":[62],"persists":[64],"throughout":[65],"RL.":[66,114,149],"Theoretically,":[67],"prove":[69],"concentration":[72],"suppresses":[73],"policy":[74],"gradient":[75],"updates":[76],"for":[77,204],"rarely":[78],"sampled":[79],"values":[81],"inflates":[83],"lower":[85],"bound":[86],"expected":[88],"Brier":[89],"risk.":[90],"To":[91,150],"overcome":[92],"exploration":[94,147],"bottleneck,":[95],"propose":[97],"CalibSFT,":[98],"plug-and-play":[100],"supervised":[101],"fine-tuning":[102],"stage":[103],"shapes":[105],"calibrated":[107],"with":[110,126],"broad":[111],"support":[112],"before":[113],"For":[115],"each":[116],"question,":[117],"CalibSFT":[118,159,183,200],"constructs":[119],"targets":[121],"combining":[122],"its":[123],"success":[124],"rate":[125],"response-level":[127],"correctness,":[128],"provably":[130],"preserves":[131],"proper-scoring":[132],"optimality,":[133],"then":[135],"balances":[136],"training":[137],"responses":[138,154,167],"across":[139,165,190],"spectrum":[142],"enable":[144],"diverse":[145],"during":[148],"learn":[151],"incorrect":[153],"without":[155],"imitating":[156],"reasoning,":[158],"introduces":[160],"correctness-conditional":[161],"supervision,":[162],"guiding":[163],"all":[166],"while":[168,195],"supervising":[169],"only":[171],"correct":[173],"ones.":[174],"Across":[175],"16":[176],"mathematical":[177],"general":[179],"benchmarks,":[181],"incorporating":[182],"reduces":[184],"calibration":[185],"errors":[186],"improves":[188],"discrimination":[189],"five":[191],"representative":[192],"RL":[193],"algorithms":[194],"preserving":[196],"comparable":[197],"accuracy.":[198],"Furthermore,":[199],"delivers":[201],"practical":[202],"benefits":[203],"downstream":[205],"selective":[206],"prediction":[207],"model":[209],"routing.":[210],"Our":[211],"code":[212],"available":[214],"at":[215],"https://github.com/ml-stat-Sustech/verbalized-confidence-training.":[216]},"counts_by_year":[{"year":2026,"cited_by_count":1}],"updated_date":"2026-11-07T11:25:07.637180","created_date":"2026-09-30T00:00:00"}