{"id":"https://openalex.org/W7130904337","doi":null,"title":"Spending Capacity Where It Matters: Selection, Adaptation, Interaction, and Structure for Efficient Language Understanding and Generation","display_name":"Spending Capacity Where It Matters: Selection, Adaptation, Interaction, and Structure for Efficient Language Understanding and Generation","publication_year":2025,"publication_date":"2025-12-01","ids":{"openalex":"https://openalex.org/W7130904337"},"language":"en","primary_location":{"id":"pmh:oai:publications.polymtl.ca:71064","doi":null,"is_oa":true,"landing_page_url":"https://publications.polymtl.ca/71064/","pdf_url":"https://publications.polymtl.ca/71064/1/2025_JonathanPilault.pdf","source":{"id":"https://openalex.org/S4306401013","display_name":"PolyPublie (\u00c9cole Polytechnique de Montr\u00e9al)","issn_l":null,"issn":null,"is_oa":false,"is_in_doaj":false,"is_core":false,"listed_in":[],"host_organization":"https://openalex.org/I45683168","host_organization_name":"Polytechnique Montr\u00e9al","host_organization_lineage":["https://openalex.org/I45683168"],"host_organization_lineage_names":[],"type":"repository"},"license":null,"license_id":null,"version":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":"Pilault, J. (2025). Spending Capacity Where It Matters: Selection, Adaptation, Interaction, and Structure for Efficient Language Understanding and Generation [Ph.D. thesis, Polytechnique Montr\u00e9al].","raw_type":"http://purl.org/coar/resource_type/c_db06"},"volume":null,"type":"dissertation","is_traditional":true,"indexed_in":[],"open_access":{"is_oa":true,"oa_status":"green","oa_url":"https://publications.polymtl.ca/71064/1/2025_JonathanPilault.pdf","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5049628803","display_name":"Jonathan Pilault","orcid":null,"observed_orcids":[]},"institutions":[{"id":"https://openalex.org/I45683168","display_name":"Polytechnique Montr\u00e9al","ror":"https://ror.org/05f8d4e86","country_code":"CA","type":"education","lineage":["https://openalex.org/I45683168"]}],"countries":["CA"],"is_corresponding":true,"raw_author_name":"Pilault, Jonathan","raw_affiliation_strings":["Polytechnique Montreal"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Polytechnique Montreal","institution_ids":["https://openalex.org/I45683168"]}]}],"institutions":[],"countries_distinct_count":1,"institutions_distinct_count":1,"corresponding_author_ids":["https://openalex.org/A5049628803"],"corresponding_institution_ids":["https://openalex.org/I45683168"],"apc_list":null,"apc_paid":null,"fwci":null,"has_fulltext":true,"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/T10028","display_name":"Topic Modeling","score":0.8339999914169312,"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":0.8339999914169312,"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/T11307","display_name":"Domain Adaptation and Few-Shot Learning","score":0.09009999781847,"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":0.05909999832510948,"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/large-language-models","display_name":"large language models","score":0.6710000038146973},{"id":"https://openalex.org/keywords/long-document-summarization","display_name":"long document summarization","score":0.859000027179718},{"id":"https://openalex.org/keywords/task-prioritization","display_name":"task prioritization","score":0.4390000104904175},{"id":"https://openalex.org/keywords/forecast-uncertainty","display_name":"forecast uncertainty","score":0.5680000185966492},{"id":"https://openalex.org/keywords/parameter-efficient-fine-tuning","display_name":"parameter-efficient fine-tuning","score":0.6930000185966492},{"id":"https://openalex.org/keywords/multi-task-learning","display_name":"multi-task learning","score":0.7239999771118164},{"id":"https://openalex.org/keywords/task-interference","display_name":"task interference","score":0.44699999690055847},{"id":"https://openalex.org/keywords/text-classification","display_name":"text classification","score":0.5260000228881836},{"id":"https://openalex.org/keywords/language-comprehension","display_name":"language comprehension","score":0.5600000023841858}],"concepts":[{"id":"https://openalex.org/C15708023","wikidata":"https://www.wikidata.org/wiki/Q80083","display_name":"Humanities","level":1,"score":0.7321000099182129},{"id":"https://openalex.org/C138885662","wikidata":"https://www.wikidata.org/wiki/Q5891","display_name":"Philosophy","level":0,"score":0.4228000044822693},{"id":"https://openalex.org/C17744445","wikidata":"https://www.wikidata.org/wiki/Q36442","display_name":"Political science","level":0,"score":0.2985000014305115},{"id":"https://openalex.org/C2779343474","wikidata":"https://www.wikidata.org/wiki/Q3109175","display_name":"Context (archaeology)","level":2,"score":0.2890999913215637},{"id":"https://openalex.org/C144024400","wikidata":"https://www.wikidata.org/wiki/Q21201","display_name":"Sociology","level":0,"score":0.2639999985694885},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.18369999527931213},{"id":"https://openalex.org/C124304363","wikidata":"https://www.wikidata.org/wiki/Q673661","display_name":"Abstraction","level":2,"score":0.1753000020980835},{"id":"https://openalex.org/C142362112","wikidata":"https://www.wikidata.org/wiki/Q735","display_name":"Art","level":0,"score":0.15119999647140503},{"id":"https://openalex.org/C136968873","wikidata":"https://www.wikidata.org/wiki/Q3181883","display_name":"Historicity (philosophy)","level":3,"score":0.1340000033378601},{"id":"https://openalex.org/C94625758","wikidata":"https://www.wikidata.org/wiki/Q7163","display_name":"Politics","level":2,"score":0.13109999895095825}],"mesh":[],"locations_count":1,"locations":[{"id":"pmh:oai:publications.polymtl.ca:71064","doi":null,"is_oa":true,"landing_page_url":"https://publications.polymtl.ca/71064/","pdf_url":"https://publications.polymtl.ca/71064/1/2025_JonathanPilault.pdf","source":{"id":"https://openalex.org/S4306401013","display_name":"PolyPublie (\u00c9cole Polytechnique de Montr\u00e9al)","issn_l":null,"issn":null,"is_oa":false,"is_in_doaj":false,"is_core":false,"listed_in":[],"host_organization":"https://openalex.org/I45683168","host_organization_name":"Polytechnique Montr\u00e9al","host_organization_lineage":["https://openalex.org/I45683168"],"host_organization_lineage_names":[],"type":"repository"},"license":null,"license_id":null,"version":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":"Pilault, J. (2025). Spending Capacity Where It Matters: Selection, Adaptation, Interaction, and Structure for Efficient Language Understanding and Generation [Ph.D. thesis, Polytechnique Montr\u00e9al].","raw_type":"http://purl.org/coar/resource_type/c_db06"}],"best_oa_location":{"id":"pmh:oai:publications.polymtl.ca:71064","doi":null,"is_oa":true,"landing_page_url":"https://publications.polymtl.ca/71064/","pdf_url":"https://publications.polymtl.ca/71064/1/2025_JonathanPilault.pdf","source":{"id":"https://openalex.org/S4306401013","display_name":"PolyPublie (\u00c9cole Polytechnique de Montr\u00e9al)","issn_l":null,"issn":null,"is_oa":false,"is_in_doaj":false,"is_core":false,"listed_in":[],"host_organization":"https://openalex.org/I45683168","host_organization_name":"Polytechnique Montr\u00e9al","host_organization_lineage":["https://openalex.org/I45683168"],"host_organization_lineage_names":[],"type":"repository"},"license":null,"license_id":null,"version":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":"Pilault, J. (2025). Spending Capacity Where It Matters: Selection, Adaptation, Interaction, and Structure for Efficient Language Understanding and Generation [Ph.D. thesis, Polytechnique Montr\u00e9al].","raw_type":"http://purl.org/coar/resource_type/c_db06"},"sustainable_development_goals":[],"study_designs":[],"awards":[],"funders":[],"has_content":{"pdf":true,"grobid_xml":true},"content_urls":{"pdf":"https://content.openalex.org/works/W7130904337.pdf","grobid_xml":"https://content.openalex.org/works/W7130904337.grobid-xml"},"referenced_works_count":87,"referenced_works":["https://openalex.org/W1816313093","https://openalex.org/W1843891098","https://openalex.org/W1997267143","https://openalex.org/W2003238113","https://openalex.org/W2117130368","https://openalex.org/W2133564696","https://openalex.org/W2307381258","https://openalex.org/W2579653291","https://openalex.org/W2609826708","https://openalex.org/W2612675303","https://openalex.org/W2626778328","https://openalex.org/W2763421725","https://openalex.org/W2767434619","https://openalex.org/W2787214294","https://openalex.org/W2793613791","https://openalex.org/W2798240283","https://openalex.org/W2799064010","https://openalex.org/W2806120502","https://openalex.org/W2809324505","https://openalex.org/W2890538051","https://openalex.org/W2896457183","https://openalex.org/W2898662126","https://openalex.org/W2898700502","https://openalex.org/W2908854766","https://openalex.org/W2911109671","https://openalex.org/W2913340405","https://openalex.org/W2913946806","https://openalex.org/W2920807444","https://openalex.org/W2922565841","https://openalex.org/W2931702774","https://openalex.org/W2940744433","https://openalex.org/W2949335953","https://openalex.org/W2950405925","https://openalex.org/W2951603207","https://openalex.org/W2952603081","https://openalex.org/W2952913664","https://openalex.org/W2953280096","https://openalex.org/W2953307569","https://openalex.org/W2955157382","https://openalex.org/W2958892036","https://openalex.org/W2962753370","https://openalex.org/W2962944953","https://openalex.org/W2962985882","https://openalex.org/W2963096510","https://openalex.org/W2963545005","https://openalex.org/W2963677766","https://openalex.org/W2963854351","https://openalex.org/W2963929190","https://openalex.org/W2963951265","https://openalex.org/W2964121744","https://openalex.org/W2964165364","https://openalex.org/W2965373594","https://openalex.org/W2980282514","https://openalex.org/W2990704537","https://openalex.org/W2997359900","https://openalex.org/W3001279689","https://openalex.org/W3022717752","https://openalex.org/W3023662336","https://openalex.org/W3027440908","https://openalex.org/W3030163527","https://openalex.org/W3033529678","https://openalex.org/W3034255912","https://openalex.org/W3034672970","https://openalex.org/W3037932933","https://openalex.org/W3040573126","https://openalex.org/W3040985171","https://openalex.org/W3091156754","https://openalex.org/W3094347401","https://openalex.org/W3095645723","https://openalex.org/W3128590981","https://openalex.org/W3129831491","https://openalex.org/W3141797743","https://openalex.org/W3203321135","https://openalex.org/W3208518372","https://openalex.org/W4224308101","https://openalex.org/W4226278401","https://openalex.org/W4226367318","https://openalex.org/W4226399820","https://openalex.org/W4281483047","https://openalex.org/W4311991106","https://openalex.org/W4317547647","https://openalex.org/W4377087886","https://openalex.org/W4385572225","https://openalex.org/W4385767747","https://openalex.org/W4399115929","https://openalex.org/W4415928588","https://openalex.org/W6967055661"],"related_works":["https://openalex.org/W2041186525","https://openalex.org/W7150168241","https://openalex.org/W7134560260","https://openalex.org/W2407749567","https://openalex.org/W7092357897","https://openalex.org/W7116638462","https://openalex.org/W4237885032","https://openalex.org/W7128129035","https://openalex.org/W4239635240","https://openalex.org/W1983778686"],"abstract_inverted_index":{"R\u00c9SUM\u00c9:":[0],"Cette":[1],"th\u00e8se":[2],"adopte":[3],"une":[4,33,80,92,96,149,301,310,338,348,355],"approche":[5],"programmatique":[6],"visant":[7],"\u00e0":[8,15,32,59,148,170,208,226,229,235,330,351,398],"amener":[9],"les":[10,155,215,233,281,335,383,425,444,447,452],"grands":[11],"mod\u00e8les":[12,234],"de":[13,54,70,82,103,108,139,162,172,190,192,194,197,285,372,375,379,385,419,427],"langage":[14],"consacrer":[16],"leur":[17],"capacit\u00e9":[18,422],"l\u00e0":[19],"o\u00f9":[20],"celle-ci":[21],"est":[22],"la":[23,55,60,101,130,145,158,209,230,246,263,320,327,405,421],"plus":[24,434],"utile.":[25],"Nous":[26],"\u00e9tudions":[27],"quatre":[28],"axes,":[29],"chacun":[30],"r\u00e9pondant":[31],"contrainte":[34],"pratique":[35],"moderne":[36],"du":[37,77,135,160,288],"traitement":[38],"automatique":[39],"des":[40,44,57,106,133,217,237,251,276,291,367,392,455],"langues,":[41],"et":[42,85,116,137,164,196,248,354,442],"proposons":[43],"conceptions":[45],"qui,":[46],"mises":[47],"ensemble,":[48],"tracent":[49],"un":[50,220,342,360],"chemin":[51],"coh\u00e9rent":[52],"allant":[53],"structuration":[56,61],"entr\u00e9es":[58],"architecturale.":[62],"Dans":[63,118,200,322],"le":[64,68,89,119,185,201,274,294,306,323,394,402,433],"Chapitre":[65,120,202,324,437],"2,":[66,508],"pour":[67,122,279,326,347,359],"r\u00e9sum\u00e9":[69],"documents":[71],"longs,":[72],"nous":[73,128,204,261,333],"pla\u00e7ons":[74],"en":[75,126,183,258,344,408,410,412,415,431],"amont":[76],"g\u00e9n\u00e9rateur":[78],"Transformer":[79,523,561],"\u00e9tape":[81],"s\u00e9lection":[83],"l\u00e9g\u00e8re":[84],"contrainte,":[86],"montrant":[87],"que":[88,100,112,378,451],"conditionnement":[90],"sur":[91,105,187,296,371,382,424],"esquisse":[93],"compacte":[94],"favorise":[95],"abstraction":[97,534],"v\u00e9ritable":[98],"plut\u00f4t":[99],"copie":[102],"surface,":[104],"\u00e9valuations":[107],"longs":[109],"textes":[110],"telles":[111],"arXiv,":[113,544],"PubMed,":[114,545],"Newsroom":[115],"BigPatent.":[117,548],"3,":[121,551],"l'apprentissage":[123],"multi-t\u00e2ches":[124,179],"\u00e9conome":[125],"param\u00e8tres,":[127],"figeons":[129],"majeure":[131],"partie":[132],"poids":[134,173],"mod\u00e8le":[136,295,343],"entra\u00eenons":[138],"petits":[140],"modules":[141],"d'adaptation":[142],"conditionn\u00e9s":[143],"par":[144,168,357],"t\u00e2che,":[146],"jumel\u00e9e":[147],"politique":[150],"d'\u00e9chantillonnage":[151],"explicite":[152],"qui":[153,212,244,271,315,430],"priorise":[154],"t\u00e2ches":[156],"selon":[157],"taille":[159],"jeu":[161],"donn\u00e9es":[163],"l'incertitude":[165,317],"pr\u00e9dictive":[166],"(pond\u00e9r\u00e9e":[167],"l'incertitude)":[169],"budget":[171],"entra\u00een\u00e9s":[174],"contraints.":[175],"Notre":[176],"adaptateur":[177],"hyperr\u00e9seau":[178],"r\u00e9duit":[180,316],"l'interf\u00e9rence":[181],"tout":[182],"am\u00e9liorant":[184],"transfert":[186],"plusieurs":[188],"jeux":[189,384],"r\u00e9f\u00e9rence":[191],"classification":[193,607],"texte":[195],"compr\u00e9hension":[198],"linguistique.":[199],"4,":[203,618],"traitons":[205],"l'ambigu\u00eft\u00e9":[206],"inh\u00e9rente":[207],"g\u00e9n\u00e9ration":[210,259,264],"linguistique":[211,329],"appara\u00eet":[213],"lorsque":[214],"requ\u00eates":[216],"utilisateurs":[218],"omettent":[219],"contexte":[221,376],"crucial.":[222],"L'absence":[223],"d'informations":[224],"propres":[225],"l'utilisateur":[227],"ou":[228,404,414],"t\u00e2che":[231],"conduit":[232],"produire":[236],"sorties":[238],"plausibles":[239],"mais":[240],"mal":[241],"align\u00e9es,":[242],"ce":[243,314],"d\u00e9grade":[245],"qualit\u00e9":[247],"provoque":[249],"souvent":[250],"reprises":[252],"co\u00fbteuses.":[253],"Au":[254,390],"moyen":[255],"d'\u00e9valuations":[256],"cibl\u00e9es":[257,278],"interlingue,":[260],"reconcevons":[262],"comme":[265],"l'\u00e9tape":[266],"finale":[267],"d'une":[268],"br\u00e8ve":[269],"interaction":[270,675],"pose,":[272],"avant":[273,401],"d\u00e9codage,":[275],"questions":[277,680],"r\u00e9v\u00e9ler":[280],"pr\u00e9f\u00e9rences":[282],"manquantes":[283],"(degr\u00e9":[284],"formalit\u00e9,":[286],"r\u00e9alisation":[287],"genre,":[289],"r\u00e9solution":[290],"pronoms).":[292],"Conditionner":[293],"ces":[297,440],"clarifications":[298,694],"l'oriente":[299],"vers":[300],"r\u00e9ponse":[302],"mieux":[303],"align\u00e9e":[304],"d\u00e8s":[305],"premier":[307],"passage,":[308],"dans":[309],"courte":[311],"phase":[312],"d'\u00e9lucidation,":[313],"sans":[318],"allonger":[319],"g\u00e9n\u00e9ration.":[321],"5,":[325,718],"mod\u00e9lisation":[328],"longue":[331],"port\u00e9e,":[332],"pr\u00e9sentons":[334],"Block-State":[336,725],"Transformers,":[337,726],"couche":[339],"hybride":[340],"combinant":[341],"espace":[345],"d'\u00e9tat,":[346],"propagation":[349,738],"efficace":[350],"long":[352,539,751],"horizon,":[353],"attention":[356,741],"blocs,":[358],"m\u00e9lange":[361],"local":[362],"sensible":[363],"au":[364],"contenu,":[365],"offrant":[366],"compromis":[368],"perplexit\u00e9-m\u00e9moire":[369],"favorables":[370],"longues":[373],"fen\u00eatres":[374],"ainsi":[377,450],"solides":[380],"performances":[381],"tests":[386],"Long":[387,757],"Range":[388,758],"Arena.":[389],"fil":[391,395],"chapitres,":[393],"conducteur":[396],"consiste":[397],"clarifier":[399],"l'essentiel":[400],"d\u00e9codage":[403],"pr\u00e9diction":[406],"-":[407,417],"s\u00e9lectionnant,":[409],"conditionnant,":[411],"interagissant":[413],"hybridant":[416],"afin":[418],"concentrer":[420],"pr\u00e9entra\u00een\u00e9e":[423],"parties":[426],"chaque":[428],"probl\u00e8me":[429],"ont":[432],"besoin.":[435],"Le":[436],"6":[438,798],"synth\u00e9tise":[439],"axes":[441],"expose":[443],"liens":[445],"avec":[446],"travaux":[448],"r\u00e9cents":[449],"prolongements":[453],"futurs":[454],"LLM":[456],"efficaces.":[457],"ABSTRACT:":[458],"This":[459],"thesis":[460],"takes":[461],"a":[462,482,497,514,522,529,591,673,707,727,732],"programmatic":[463],"view":[464],"of":[465,521,559,672,790,811],"getting":[466],"large":[467],"language":[468,609,625,721],"models":[469,641,734],"to":[470,503,642,681,767,805],"spend":[471],"capacity":[472,784],"where":[473],"it":[474,795],"matters.":[475],"We":[476],"study":[477],"four":[478],"settings":[479],"each":[480,791],"addressing":[481],"practical":[483],"pressure":[484],"in":[485,519,624],"modern":[486],"Natural":[487],"Language":[488],"Processing":[489],"and":[490,547,564,586,608,651,753,802,808],"report":[491],"designs":[492],"that,":[493],"taken":[494],"together,":[495],"form":[496],"coherent":[498],"path":[499],"from":[500],"input":[501],"structuring":[502],"architectural":[504],"structuring.":[505],"In":[506,549,616,716],"Chapter":[507,550,617,717,797],"for":[509,552,719,735,742],"long-document":[510],"summarization":[511],"we":[512,556,619,665,723],"place":[513],"lightweight,":[515],"constrained":[516],"selection":[517],"step":[518,671],"front":[520],"generator,":[524],"showing":[525],"that":[526,579,627,676,730,782,793],"conditioning":[527],"on":[528,604,662,692,702,756,787],"compact":[530],"sketch":[531],"encourages":[532],"genuine":[533],"over":[535],"surface":[536,682],"copying":[537],"across":[538],"text":[540,606],"evaluations":[541,661],"such":[542],"as":[543,668],"Newsroom,":[546],"parameter-efficient":[553],"multi-task":[554,596],"learning":[555],"keep":[557],"most":[558],"the":[560,621,669,683,696,699,703,763,788],"layers":[562],"frozen":[563],"route":[565],"task":[566],"variation":[567],"through":[568],"small,":[569],"task-conditioned":[570],"adapter":[571,598],"modules,":[572],"together":[573],"with":[574,612,739],"an":[575],"explicit":[576],"sampling":[577],"policy":[578],"prioritizes":[580],"tasks":[581],"by":[582,775],"dataset":[583],"size":[584],"(temperature-scaled)":[585],"predictive":[587],"uncertainty":[588,712],"(uncertainty-weighted)":[589],"under":[590],"fixed":[592],"trained-parameter":[593],"budget.":[594],"Our":[595],"hypernetwork":[597],"reduces":[599],"interference":[600],"while":[601],"improving":[602],"transfer":[603],"several":[605],"understanding":[610],"benchmarks":[611],"minimal":[613],"added":[614],"capacity.":[615],"address":[620],"inherent":[622],"ambiguity":[623],"generation":[626,667],"arises":[628],"when":[629],"user":[630],"queries":[631],"omit":[632],"crucial":[633],"context.":[634],"Missing":[635],"user-specific":[636],"or":[637,656,773,779],"task-specific":[638],"information":[639],"leads":[640],"produce":[643],"plausible":[644],"yet":[645],"misaligned":[646],"outputs,":[647],"which":[648],"degrades":[649],"quality":[650],"often":[652],"triggers":[653],"costly":[654],"retries":[655],"longer":[657],"prompts.":[658],"With":[659],"specific":[660],"cross-lingual":[663],"generation,":[664],"recast":[666],"final":[670],"brief":[674,708],"asks":[677],"targeted":[678],"pre-decoding":[679],"missing":[684],"preferences":[685],"(formality,":[686],"gender":[687],"realization,":[688],"pronoun":[689],"resolution).":[690],"Conditioning":[691],"these":[693,800],"steers":[695],"model":[697],"toward":[698],"intended":[700],"output":[701],"first":[704],"pass,":[705],"within":[706],"elicitation":[709],"phase,":[710],"reducing":[711],"without":[713],"lengthening":[714],"generation.":[715],"long-range":[720],"modeling":[722],"introduce":[724],"hybrid":[728],"layer":[729],"composes":[731],"state-space":[733],"efficient":[736,812],"long-horizon":[737],"block-wise":[740],"local,":[743],"content-aware":[744],"mixing,":[745],"yielding":[746],"favorable":[747],"perplexity\u2013memory":[748],"trade-offs":[749],"at":[750],"context":[752],"strong":[754],"performance":[755],"Arena":[759],"benchmarks.":[760],"Across":[761],"chapters,":[762],"unifying":[764],"theme":[765],"is":[766,785],"clarify":[768],"what":[769],"matters":[770],"before":[771],"decoding":[772],"prediction":[774],"selecting,":[776],"conditioning,":[777],"interacting,":[778],"hybridizing":[780],"so":[781],"pretrained":[783],"focused":[786],"parts":[789],"problem":[792],"need":[794],"most.":[796],"synthesizes":[799],"threads":[801],"outlines":[803],"links":[804],"current":[806],"works":[807],"future":[809],"extensions":[810],"LLMs.":[813]},"counts_by_year":[],"updated_date":"2026-11-07T11:25:07.637180","created_date":"2026-02-22T00:00:00"}