{"id":"https://openalex.org/W7160613116","doi":"https://doi.org/10.48550/arxiv.2605.06294","title":"Log-Likelihood, Simpson's Paradox, and the Detection of Machine-Generated Text","display_name":"Log-Likelihood, Simpson's Paradox, and the Detection of Machine-Generated Text","publication_year":2026,"publication_date":"2026-05-07","ids":{"openalex":"https://openalex.org/W7160613116","doi":"https://doi.org/10.48550/arxiv.2605.06294"},"language":null,"primary_location":{"id":"doi:10.48550/arxiv.2605.06294","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2605.06294","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":"Preprint"},"type":"preprint","indexed_in":["datacite"],"open_access":{"is_oa":true,"oa_status":"green","oa_url":"https://doi.org/10.48550/arxiv.2605.06294","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5083538214","display_name":"Tom Kempton","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Kempton, Tom","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5135669393","display_name":"Viktor Drobnyi","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Drobnyi, Viktor","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5047161303","display_name":"Maeve Madigan","orcid":"https://orcid.org/0000-0003-2485-8960"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Madigan, Maeve","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5125889107","display_name":"Stuart Burrell","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Burrell, Stuart","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":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/T11689","display_name":"Adversarial Robustness in Machine Learning","score":0.12950000166893005,"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":0.12950000166893005,"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":0.08709999918937683,"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/T10775","display_name":"Generative Adversarial Networks and Image Synthesis","score":0.07289999723434448,"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"}}],"keywords":[{"id":"https://openalex.org/keywords/detector","display_name":"Detector","score":0.695900022983551},{"id":"https://openalex.org/keywords/security-token","display_name":"Security token","score":0.6051999926567078},{"id":"https://openalex.org/keywords/bayesian-probability","display_name":"Bayesian probability","score":0.5698999762535095},{"id":"https://openalex.org/keywords/aggregate","display_name":"Aggregate (composite)","score":0.5658000111579895},{"id":"https://openalex.org/keywords/calibration","display_name":"Calibration","score":0.5622000098228455},{"id":"https://openalex.org/keywords/natural-language","display_name":"Natural language","score":0.5440000295639038},{"id":"https://openalex.org/keywords/signal","display_name":"SIGNAL (programming language)","score":0.44620001316070557},{"id":"https://openalex.org/keywords/exploit","display_name":"Exploit","score":0.41620001196861267},{"id":"https://openalex.org/keywords/modular-design","display_name":"Modular design","score":0.38940000534057617}],"concepts":[{"id":"https://openalex.org/C94915269","wikidata":"https://www.wikidata.org/wiki/Q1834857","display_name":"Detector","level":2,"score":0.695900022983551},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.6593999862670898},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.6302000284194946},{"id":"https://openalex.org/C48145219","wikidata":"https://www.wikidata.org/wiki/Q1335365","display_name":"Security token","level":2,"score":0.6051999926567078},{"id":"https://openalex.org/C107673813","wikidata":"https://www.wikidata.org/wiki/Q812534","display_name":"Bayesian probability","level":2,"score":0.5698999762535095},{"id":"https://openalex.org/C4679612","wikidata":"https://www.wikidata.org/wiki/Q866298","display_name":"Aggregate (composite)","level":2,"score":0.5658000111579895},{"id":"https://openalex.org/C165838908","wikidata":"https://www.wikidata.org/wiki/Q736777","display_name":"Calibration","level":2,"score":0.5622000098228455},{"id":"https://openalex.org/C195324797","wikidata":"https://www.wikidata.org/wiki/Q33742","display_name":"Natural language","level":2,"score":0.5440000295639038},{"id":"https://openalex.org/C119857082","wikidata":"https://www.wikidata.org/wiki/Q2539","display_name":"Machine learning","level":1,"score":0.4771000146865845},{"id":"https://openalex.org/C2779843651","wikidata":"https://www.wikidata.org/wiki/Q7390335","display_name":"SIGNAL (programming language)","level":2,"score":0.44620001316070557},{"id":"https://openalex.org/C165696696","wikidata":"https://www.wikidata.org/wiki/Q11287","display_name":"Exploit","level":2,"score":0.41620001196861267},{"id":"https://openalex.org/C101468663","wikidata":"https://www.wikidata.org/wiki/Q1620158","display_name":"Modular design","level":2,"score":0.38940000534057617},{"id":"https://openalex.org/C2778572836","wikidata":"https://www.wikidata.org/wiki/Q380933","display_name":"Space (punctuation)","level":2,"score":0.38089999556541443},{"id":"https://openalex.org/C204321447","wikidata":"https://www.wikidata.org/wiki/Q30642","display_name":"Natural language processing","level":1,"score":0.3799000084400177},{"id":"https://openalex.org/C44291984","wikidata":"https://www.wikidata.org/wiki/Q1074173","display_name":"Question answering","level":2,"score":0.3634999990463257},{"id":"https://openalex.org/C160234255","wikidata":"https://www.wikidata.org/wiki/Q812535","display_name":"Bayesian inference","level":3,"score":0.3531000018119812},{"id":"https://openalex.org/C137293760","wikidata":"https://www.wikidata.org/wiki/Q3621696","display_name":"Language model","level":2,"score":0.3449000120162964},{"id":"https://openalex.org/C87007009","wikidata":"https://www.wikidata.org/wiki/Q210832","display_name":"Statistical hypothesis testing","level":2,"score":0.34040001034736633},{"id":"https://openalex.org/C12725497","wikidata":"https://www.wikidata.org/wiki/Q810247","display_name":"Baseline (sea)","level":2,"score":0.3402999937534332},{"id":"https://openalex.org/C153180895","wikidata":"https://www.wikidata.org/wiki/Q7148389","display_name":"Pattern recognition (psychology)","level":2,"score":0.3379000127315521},{"id":"https://openalex.org/C198082294","wikidata":"https://www.wikidata.org/wiki/Q3399648","display_name":"Position (finance)","level":2,"score":0.3287000060081482},{"id":"https://openalex.org/C114289077","wikidata":"https://www.wikidata.org/wiki/Q3284399","display_name":"Statistical model","level":2,"score":0.3197999894618988},{"id":"https://openalex.org/C115988155","wikidata":"https://www.wikidata.org/wiki/Q3262192","display_name":"Decision problem","level":2,"score":0.28130000829696655},{"id":"https://openalex.org/C184337299","wikidata":"https://www.wikidata.org/wiki/Q1437428","display_name":"Semantics (computer science)","level":2,"score":0.27480000257492065},{"id":"https://openalex.org/C28901747","wikidata":"https://www.wikidata.org/wiki/Q177571","display_name":"Decision theory","level":2,"score":0.25690001249313354},{"id":"https://openalex.org/C124101348","wikidata":"https://www.wikidata.org/wiki/Q172491","display_name":"Data mining","level":1,"score":0.2565999925136566},{"id":"https://openalex.org/C158154518","wikidata":"https://www.wikidata.org/wiki/Q7310970","display_name":"Relevance (law)","level":2,"score":0.2565999925136566},{"id":"https://openalex.org/C26517878","wikidata":"https://www.wikidata.org/wiki/Q228039","display_name":"Key (lock)","level":2,"score":0.25519999861717224},{"id":"https://openalex.org/C137270730","wikidata":"https://www.wikidata.org/wiki/Q120811","display_name":"Detection theory","level":3,"score":0.2529999911785126},{"id":"https://openalex.org/C2776187449","wikidata":"https://www.wikidata.org/wiki/Q1513879","display_name":"Natural language generation","level":3,"score":0.25220000743865967}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.48550/arxiv.2605.06294","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2605.06294","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":"Preprint"}],"best_oa_location":{"id":"doi:10.48550/arxiv.2605.06294","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2605.06294","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":"Preprint"},"sustainable_development_goals":[{"id":"https://metadata.un.org/sdg/16","display_name":"Peace, Justice and strong institutions","score":0.5504897236824036}],"awards":[],"funders":[],"has_content":{"pdf":false,"grobid_xml":false},"content_urls":null,"referenced_works_count":0,"referenced_works":[],"related_works":[],"abstract_inverted_index":{"The":[0,19],"ability":[1],"to":[2,22,36,170,230],"reliably":[3],"distinguish":[4],"human-written":[5,42],"text":[6,31,55],"from":[7,168],"that":[8,29,47],"generated":[9],"by":[10,95],"large":[11],"language":[12,39],"models":[13],"is":[14,56,93,193],"of":[15,62,86,125,164,202,206,208],"profound":[16],"societal":[17],"importance.":[18],"dominant":[20],"approach":[21],"this":[23],"problem":[24],"exploits":[25],"the":[26,48,59,63,126,186,228,251],"likelihood":[27],"hypothesis:":[28],"machine-generated":[30],"should":[32],"appear":[33],"more":[34],"probable":[35],"a":[37,84,89,104,176,195,199,203,212,225],"detector":[38,64,179],"model":[40],"than":[41,115],"text.":[43],"However,":[44],"we":[45,102,120,157,180],"demonstrate":[46],"token-level":[49],"signal":[50,92],"distinguishing":[51],"human":[52],"and":[53,66,135,145,154,175,211,243],"machine":[54],"non-uniform":[57],"across":[58,72,150,185],"hidden":[60,133],"space":[61],"model,":[65],"naively":[67],"averaging":[68],"likelihood-based":[69],"token":[70,118],"scores":[71],"regions":[73],"with":[74,217,233,246],"fundamentally":[75],"different":[76],"statistical":[77],"structure,":[78],"as":[79,224],"most":[80],"detectors":[81,153,210],"do,":[82],"causes":[83],"form":[85],"Simpson's":[87],"paradox:":[88],"strong":[90],"local":[91,106],"destroyed":[94],"inappropriate":[96],"aggregation.":[97],"To":[98],"correct":[99],"for":[100,227],"this,":[101],"introduce":[103,181],"learned":[105],"calibration":[107,241],"step":[108],"grounded":[109],"in":[110,132],"Bayesian":[111],"decision":[112,254],"theory.":[113],"Rather":[114],"aggregating":[116],"raw":[117],"scores,":[119],"first":[121],"learn":[122],"lightweight":[123],"predictors":[124],"score":[127],"distributions":[128],"conditioned":[129],"on":[130,172],"position":[131],"space,":[134],"aggregate":[136],"calibrated":[137,162],"log-likelihood":[138],"ratios":[139],"instead.":[140],"This":[141,221],"single":[142],"intervention":[143],"dramatically":[144],"consistently":[146],"improves":[147,166],"detection":[148],"performance":[149,184],"all":[151,155],"baseline":[152],"datasets":[156],"consider.":[158],"For":[159],"example,":[160],"our":[161,190],"variant":[163],"Fast-DetectGPT":[165],"AUROC":[167],"$0.63$":[169],"$0.85$":[171],"GPT-5.4":[173],"text,":[174],"locally-calibrated":[177],"DMAP":[178],"achieves":[182],"state-of-the-art":[183],"board.":[187],"That":[188],"said,":[189],"central":[191],"contribution":[192],"not":[194],"new":[196],"detector,":[197],"but":[198],"precise":[200],"diagnosis":[201],"significant":[204],"cause":[205],"under-performance":[207],"existing":[209],"principled,":[213],"modular":[214],"remedy":[215],"compatible":[216],"any":[218],"token-averaging":[219],"pipeline.":[220],"will":[222],"serve":[223],"foundation":[226],"community":[229],"build":[231],"upon,":[232],"natural":[234],"avenues":[235],"including":[236],"richer":[237],"distributional":[238],"models,":[239],"improved":[240],"strategies,":[242],"principled":[244],"ensembling":[245],"hidden-space":[247],"geometry":[248],"signals":[249],"via":[250],"full":[252],"Bayes-optimal":[253],"rule.":[255]},"counts_by_year":[],"updated_date":"2026-07-01T06:00:48.157686","created_date":"2026-05-09T00:00:00"}