{"id":"https://openalex.org/W7140191897","doi":"https://doi.org/10.48550/arxiv.2603.20531","title":"Epistemic Observability in Language Models","display_name":"Epistemic Observability in Language Models","publication_year":2026,"publication_date":"2026-03-20","ids":{"openalex":"https://openalex.org/W7140191897","doi":"https://doi.org/10.48550/arxiv.2603.20531"},"language":"en","primary_location":{"id":"pmh:oai:arXiv.org:2603.20531","doi":null,"is_oa":true,"landing_page_url":"https://arxiv.org/abs/2603.20531","pdf_url":"https://arxiv.org/pdf/2603.20531","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","is_traditional":true,"indexed_in":["arxiv","datacite"],"open_access":{"is_oa":true,"oa_status":"green","oa_url":"https://arxiv.org/pdf/2603.20531","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5149688844","display_name":"Tony Mason","orcid":null,"observed_orcids":[]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Mason, Tony","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5050938124","display_name":"Vaastav Anand","orcid":"https://orcid.org/0000-0001-8502-0657","observed_orcids":["https://orcid.org/0000-0001-8502-0657"]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Anand, Vaastav","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":2,"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":0.6338000297546387,"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":0.6338000297546387,"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":0.08980000019073486,"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/T10883","display_name":"Ethics and Social Impacts of AI","score":0.05130000039935112,"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/epistemic-honesty","display_name":"epistemic honesty","score":0.8550000190734863},{"id":"https://openalex.org/keywords/hallucination-detection","display_name":"hallucination detection","score":0.6840000152587891},{"id":"https://openalex.org/keywords/impossibility-results","display_name":"impossibility results","score":0.7039999961853027},{"id":"https://openalex.org/keywords/reinforcement-learning-from-human-feedback","display_name":"reinforcement learning from human feedback","score":0.4860000014305115},{"id":"https://openalex.org/keywords/instruction-tuning","display_name":"instruction tuning","score":0.9610000252723694}],"concepts":[{"id":"https://openalex.org/C55439883","wikidata":"https://www.wikidata.org/wiki/Q360812","display_name":"Correctness","level":2,"score":0.767799973487854},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.6401000022888184},{"id":"https://openalex.org/C36299963","wikidata":"https://www.wikidata.org/wiki/Q1369844","display_name":"Observability","level":2,"score":0.6388999819755554},{"id":"https://openalex.org/C2776261394","wikidata":"https://www.wikidata.org/wiki/Q315562","display_name":"Impossibility","level":2,"score":0.5486999750137329},{"id":"https://openalex.org/C137293760","wikidata":"https://www.wikidata.org/wiki/Q3621696","display_name":"Language model","level":2,"score":0.527999997138977},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.4429999887943268},{"id":"https://openalex.org/C106301342","wikidata":"https://www.wikidata.org/wiki/Q4117933","display_name":"Entropy (arrow of time)","level":2,"score":0.4415999948978424},{"id":"https://openalex.org/C2779110517","wikidata":"https://www.wikidata.org/wiki/Q1240788","display_name":"Supervisor","level":2,"score":0.4399999976158142},{"id":"https://openalex.org/C80444323","wikidata":"https://www.wikidata.org/wiki/Q2878974","display_name":"Theoretical computer science","level":1,"score":0.4226999878883362},{"id":"https://openalex.org/C119857082","wikidata":"https://www.wikidata.org/wiki/Q2539","display_name":"Machine learning","level":1,"score":0.382099986076355},{"id":"https://openalex.org/C44291984","wikidata":"https://www.wikidata.org/wiki/Q1074173","display_name":"Question answering","level":2,"score":0.38119998574256897},{"id":"https://openalex.org/C93959086","wikidata":"https://www.wikidata.org/wiki/Q6888345","display_name":"Model selection","level":2,"score":0.37139999866485596},{"id":"https://openalex.org/C121649496","wikidata":"https://www.wikidata.org/wiki/Q14402006","display_name":"Reductio ad absurdum","level":3,"score":0.36719998717308044},{"id":"https://openalex.org/C9679016","wikidata":"https://www.wikidata.org/wiki/Q1417473","display_name":"Principle of maximum entropy","level":2,"score":0.36469998955726624},{"id":"https://openalex.org/C2164484","wikidata":"https://www.wikidata.org/wiki/Q5170150","display_name":"Core (optical fiber)","level":2,"score":0.3474999964237213},{"id":"https://openalex.org/C2780704645","wikidata":"https://www.wikidata.org/wiki/Q9251458","display_name":"Observer (physics)","level":2,"score":0.33009999990463257},{"id":"https://openalex.org/C204321447","wikidata":"https://www.wikidata.org/wiki/Q30642","display_name":"Natural language processing","level":1,"score":0.3084999918937683},{"id":"https://openalex.org/C2780801425","wikidata":"https://www.wikidata.org/wiki/Q5164392","display_name":"Construct (python library)","level":2,"score":0.30410000681877136},{"id":"https://openalex.org/C8505890","wikidata":"https://www.wikidata.org/wiki/Q605095","display_name":"Budget constraint","level":2,"score":0.30329999327659607},{"id":"https://openalex.org/C191399111","wikidata":"https://www.wikidata.org/wiki/Q64861","display_name":"Transitive relation","level":2,"score":0.29510000348091125},{"id":"https://openalex.org/C11413529","wikidata":"https://www.wikidata.org/wiki/Q8366","display_name":"Algorithm","level":1,"score":0.29170000553131104},{"id":"https://openalex.org/C48044578","wikidata":"https://www.wikidata.org/wiki/Q727490","display_name":"Scalability","level":2,"score":0.27900001406669617},{"id":"https://openalex.org/C120936955","wikidata":"https://www.wikidata.org/wiki/Q2155640","display_name":"Empirical research","level":2,"score":0.27869999408721924},{"id":"https://openalex.org/C104824951","wikidata":"https://www.wikidata.org/wiki/Q830124","display_name":"Girsanov theorem","level":3,"score":0.27410000562667847},{"id":"https://openalex.org/C195324797","wikidata":"https://www.wikidata.org/wiki/Q33742","display_name":"Natural language","level":2,"score":0.2669000029563904},{"id":"https://openalex.org/C44249647","wikidata":"https://www.wikidata.org/wiki/Q208498","display_name":"Confidence interval","level":2,"score":0.26260000467300415},{"id":"https://openalex.org/C152442038","wikidata":"https://www.wikidata.org/wiki/Q2778212","display_name":"Tikhonov regularization","level":3,"score":0.2581999897956848},{"id":"https://openalex.org/C110251889","wikidata":"https://www.wikidata.org/wiki/Q1569697","display_name":"Model checking","level":2,"score":0.2558000087738037},{"id":"https://openalex.org/C2778755073","wikidata":"https://www.wikidata.org/wiki/Q10858537","display_name":"Scale (ratio)","level":2,"score":0.25459998846054077},{"id":"https://openalex.org/C179799912","wikidata":"https://www.wikidata.org/wiki/Q205084","display_name":"Computational complexity theory","level":2,"score":0.2531000077724457}],"mesh":[],"locations_count":2,"locations":[{"id":"pmh:oai:arXiv.org:2603.20531","doi":null,"is_oa":true,"landing_page_url":"https://arxiv.org/abs/2603.20531","pdf_url":"https://arxiv.org/pdf/2603.20531","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.2603.20531","doi":"https://doi.org/10.48550/arxiv.2603.20531","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2603.20531","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:2603.20531","doi":null,"is_oa":true,"landing_page_url":"https://arxiv.org/abs/2603.20531","pdf_url":"https://arxiv.org/pdf/2603.20531","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/W7140191897.pdf","grobid_xml":"https://content.openalex.org/works/W7140191897.grobid-xml"},"referenced_works_count":11,"referenced_works":["https://openalex.org/W1985787415","https://openalex.org/W4223908421","https://openalex.org/W4285429195","https://openalex.org/W4361012985","https://openalex.org/W4387561528","https://openalex.org/W4401306886","https://openalex.org/W4414939355","https://openalex.org/W4417231378","https://openalex.org/W6948229050","https://openalex.org/W7116043699","https://openalex.org/W7124455761"],"related_works":["https://openalex.org/W7160418535","https://openalex.org/W7167686704","https://openalex.org/W7171678872","https://openalex.org/W7223060502","https://openalex.org/W7166883315","https://openalex.org/W7167473530","https://openalex.org/W7212045387","https://openalex.org/W4416716572","https://openalex.org/W7203935888","https://openalex.org/W7133213730"],"abstract_inverted_index":{"We":[0,38,79,147],"find":[1],"that":[2,44,84,101,152,165],"models":[3],"report":[4],"highest":[5],"confidence":[6,21],"precisely":[7],"when":[8,116],"they":[9],"are":[10,120,166,256],"fabricating.":[11],"Across":[12],"four":[13],"model":[14,74,137],"families":[15],"(OLMo-3,":[16],"Llama-3.1,":[17],"Qwen3,":[18],"Mistral),":[19],"self-reported":[20],"inversely":[22],"correlates":[23],"with":[24,26],"accuracy,":[25],"AUC":[27,178],"ranging":[28],"from":[29,76,107,216],"0.28":[30],"to":[31,113,169,225,241],"0.36":[32],"where":[33,58,212],"0.5":[34],"is":[35,46,208,232,247,252],"random":[36],"guessing.":[37],"prove,":[39],"under":[40,171],"explicit":[41],"formal":[42,130],"assumptions,":[43],"this":[45],"not":[47],"a":[48,59,108,149,209,233],"capability":[49],"gap":[50],"but":[51],"an":[52],"observational":[53],"one.":[54],"Under":[55],"text-only":[56,109],"observation,":[57],"supervisor":[60,110],"sees":[61],"only":[62,88],"the":[63,90,117,154,213,248,253],"model's":[64],"output":[65],"text,":[66],"no":[67,102],"monitoring":[68],"system":[69,237,254],"can":[70,111],"reliably":[71],"distinguish":[72],"honest":[73,114],"outputs":[75],"plausible":[77],"fabrications.":[78],"prove":[80],"two":[81],"results:":[82],"first,":[83],"any":[85],"policy":[86],"conditioning":[87],"on":[89],"query":[91],"cannot":[92],"satisfy":[93],"epistemic":[94],"honesty":[95],"across":[96,200],"ambiguous":[97],"world":[98],"states;":[99],"second,":[100],"learning":[103],"algorithm":[104],"optimizing":[105],"reward":[106],"converge":[112],"behavior":[115],"supervisor's":[118],"observations":[119],"identical":[121],"for":[122,228,236],"both":[123],"grounded":[124],"and":[125,144,162],"fabricated":[126],"responses.":[127],"Within":[128],"our":[129],"model,":[131],"these":[132],"impossibilities":[133],"hold":[134],"regardless":[135],"of":[136,220],"scale":[138],"or":[139],"training":[140],"procedure,":[141],"including":[142],"RLHF":[143],"instruction":[145],"tuning.":[146],"construct":[148],"tensor":[150],"interface":[151],"escapes":[153],"impossibility":[155],"by":[156,184],"exporting":[157],"computational":[158],"byproducts":[159],"(per-token":[160],"entropy":[161,175,197],"log-probability":[163],"distributions)":[164],"structurally":[167],"coupled":[168],"correctness":[170],"standard":[172],"training.":[173],"Per-token":[174],"achieves":[176],"pooled":[177],"0.757,":[179],"outperforming":[180],"all":[181],"text":[182],"baselines":[183],"2.5--3.9":[185],"percentage":[186],"points":[187],"at":[188],"every":[189],"budget":[190,218],"level":[191],"tested":[192],"(10\\%,":[193],"20\\%,":[194],"30\\%).":[195],"The":[196,205,245,250],"signal":[198],"generalizes":[199],"architectures":[201],"(Spearman":[202],"$\u03c1=":[203],"0.762$).":[204],"core":[206],"contribution":[207,246],"cost":[210],"surface":[211],"empirical":[214],"mapping":[215],"verification":[217,243],"(fraction":[219],"queries":[221],"receiving":[222],"expensive":[223],"checks)":[224],"detection":[226],"accuracy":[227],"each":[229],"judge":[230],"strategy":[231],"practical":[234],"lookup":[235],"builders":[238],"deciding":[239],"how":[240],"allocate":[242],"resources.":[244],"map.":[249],"territory":[251],"you":[255],"building.":[257]},"counts_by_year":[{"year":2026,"cited_by_count":1},{"year":2025,"cited_by_count":1}],"updated_date":"2026-11-07T11:25:07.637180","created_date":"2026-03-25T00:00:00"}