{"id":"https://openalex.org/W4406548552","doi":"https://doi.org/10.48550/arxiv.2501.09345","title":"Rational Tuning of LLM Cascades via Probabilistic Modeling","display_name":"Rational Tuning of LLM Cascades via Probabilistic Modeling","publication_year":2025,"publication_date":"2025-01-16","ids":{"openalex":"https://openalex.org/W4406548552","doi":"https://doi.org/10.48550/arxiv.2501.09345"},"language":"en","primary_location":{"id":"pmh:oai:arXiv.org:2501.09345","doi":null,"is_oa":true,"landing_page_url":"http://arxiv.org/abs/2501.09345","pdf_url":"https://arxiv.org/pdf/2501.09345","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/2501.09345","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5114452247","display_name":"Michael J. Zellinger","orcid":null,"observed_orcids":[]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Zellinger, Michael J.","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5080587799","display_name":"Matt Thomson","orcid":null,"observed_orcids":[]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Thomson, Matt","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":5,"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":0.5669000148773193,"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":0.5669000148773193,"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/T10181","display_name":"Natural Language Processing Techniques","score":0.22059999406337738,"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/T11689","display_name":"Adversarial Robustness in Machine Learning","score":0.13570000231266022,"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/stochastic-modeling","display_name":"stochastic modeling","score":0.6930000185966492},{"id":"https://openalex.org/keywords/hallucinations","display_name":"hallucinations","score":0.722000002861023},{"id":"https://openalex.org/keywords/sample-efficiency","display_name":"sample efficiency","score":0.746999979019165},{"id":"https://openalex.org/keywords/bayesian-optimization-comparison","display_name":"Bayesian optimization comparison","score":0.6359999775886536}],"concepts":[{"id":"https://openalex.org/C49937458","wikidata":"https://www.wikidata.org/wiki/Q2599292","display_name":"Probabilistic logic","level":2,"score":0.6317119002342224},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.5390619039535522},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.33731287717819214}],"mesh":[],"locations_count":2,"locations":[{"id":"pmh:oai:arXiv.org:2501.09345","doi":null,"is_oa":true,"landing_page_url":"http://arxiv.org/abs/2501.09345","pdf_url":"https://arxiv.org/pdf/2501.09345","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.2501.09345","doi":"https://doi.org/10.48550/arxiv.2501.09345","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2501.09345","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":null,"license_id":null,"version":null,"is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"Preprint"}],"best_oa_location":{"id":"pmh:oai:arXiv.org:2501.09345","doi":null,"is_oa":true,"landing_page_url":"http://arxiv.org/abs/2501.09345","pdf_url":"https://arxiv.org/pdf/2501.09345","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/W4406548552.pdf","grobid_xml":"https://content.openalex.org/works/W4406548552.grobid-xml"},"referenced_works_count":1,"referenced_works":["https://openalex.org/W3210904070"],"related_works":["https://openalex.org/W7171664955","https://openalex.org/W7160629695","https://openalex.org/W7212012822","https://openalex.org/W4387427491","https://openalex.org/W4416037319","https://openalex.org/W7161700234","https://openalex.org/W4416159353","https://openalex.org/W7154655983","https://openalex.org/W4403570545","https://openalex.org/W4415428215"],"abstract_inverted_index":{"Understanding":[0],"the":[1,33,40,64,80,96,124,127,141,150,163,166,191],"reliability":[2],"of":[3,35,67,84,87,99,169,193,199],"large":[4],"language":[5],"models":[6,69],"(LLMs)":[7],"has":[8],"recently":[9],"garnered":[10],"significant":[11],"attention.":[12],"Given":[13],"LLMs'":[14],"propensity":[15],"to":[16,24,31,55,107,154,190],"hallucinate,":[17],"as":[18,20,50],"well":[19],"their":[21],"high":[22],"sensitivity":[23],"prompt":[25],"design,":[26],"it":[27],"is":[28],"already":[29],"challenging":[30],"predict":[32],"performance":[34,82,151,187],"an":[36],"individual":[37],"LLM.":[38],"However,":[39],"problem":[41],"becomes":[42],"more":[43,119],"complex":[44],"for":[45,79,93,134,183],"compound":[46],"LLM":[47,101,185],"systems":[48,198],"such":[49],"cascades,":[51],"where":[52],"in":[53,196],"addition":[54],"each":[56],"model's":[57],"standalone":[58],"performance,":[59],"we":[60,74],"must":[61],"understand":[62],"how":[63],"error":[65,167],"rates":[66,168],"different":[68,170],"interact.":[70],"In":[71,140],"this":[72],"paper,":[73],"present":[75],"a":[76,85,91,100,180],"probabilistic":[77,194],"model":[78,117,178],"joint":[81],"distribution":[83],"sequence":[86],"LLMs,":[88],"which":[89],"enables":[90],"framework":[92],"rationally":[94],"tuning":[95,184],"confidence":[97,109],"thresholds":[98,110],"cascade":[102,186],"using":[103,111],"continuous":[104],"optimization.":[105],"Compared":[106],"selecting":[108],"Bayesian":[112],"optimization,":[113],"our":[114,158,176],"parametric":[115],"Markov-copula":[116,177],"yields":[118],"favorable":[120],"error-cost":[121,128],"trade-offs,":[122],"improving":[123],"area":[125],"under":[126],"curve":[129],"by":[130],"4.3%":[131],"on":[132],"average":[133],"cascades":[135],"with":[136,144],"$k\\geq":[137],"3$":[138],"models.":[139],"low-sample":[142],"regime":[143],"$n":[145],"\\leq":[146],"30$":[147],"training":[148],"examples,":[149],"improvement":[152],"widens":[153],"10.2%,":[155],"suggesting":[156],"that":[157],"framework's":[159],"inductive":[160],"assumptions":[161],"about":[162],"interactions":[164],"between":[165],"LLMs":[171],"enhance":[172],"sample":[173],"efficiency.":[174],"Overall,":[175],"provides":[179],"rational":[181],"basis":[182],"and":[188],"points":[189],"potential":[192],"methods":[195],"analyzing":[197],"LLMs.":[200]},"counts_by_year":[{"year":2026,"cited_by_count":3},{"year":2025,"cited_by_count":2}],"updated_date":"2026-11-07T11:25:07.637180","created_date":"2025-01-18T00:00:00"}