{"id":"https://openalex.org/W4392737513","doi":"https://doi.org/10.48550/arxiv.2403.06963","title":"The pitfalls of next-token prediction","display_name":"The pitfalls of next-token prediction","publication_year":2024,"publication_date":"2024-03-11","ids":{"openalex":"https://openalex.org/W4392737513","doi":"https://doi.org/10.48550/arxiv.2403.06963"},"language":"en","primary_location":{"id":"pmh:oai:arXiv.org:2403.06963","doi":null,"is_oa":true,"landing_page_url":"http://arxiv.org/abs/2403.06963","pdf_url":"https://arxiv.org/pdf/2403.06963","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","indexed_in":["arxiv","datacite"],"open_access":{"is_oa":true,"oa_status":"green","oa_url":"https://arxiv.org/pdf/2403.06963","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5034212403","display_name":"Gregor Bachmann","orcid":"https://orcid.org/0000-0002-7682-6381","observed_orcids":["https://orcid.org/0000-0002-7682-6381"]},"institutions":[{"id":"https://openalex.org/I35440088","display_name":"ETH Zurich","ror":"https://ror.org/05a28rw58","country_code":"CH","type":"education","lineage":["https://openalex.org/I2799323385","https://openalex.org/I35440088"]}],"countries":["CH"],"is_corresponding":false,"raw_author_name":"Bachmann, Gregor","raw_affiliation_strings":["ETH Z\u00fcrich , Switzerland","Equal contribution"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"ETH Z\u00fcrich , Switzerland","institution_ids":["https://openalex.org/I35440088"]},{"raw_affiliation_string":"Equal contribution","institution_ids":[]}]},{"author_position":"last","author":{"id":"https://openalex.org/A5108253513","display_name":"Vaishnavh Nagarajan","orcid":null,"observed_orcids":[]},"institutions":[{"id":"https://openalex.org/I1291425158","display_name":"Google (United States)","ror":"https://ror.org/00njsd438","country_code":"US","type":"company","lineage":["https://openalex.org/I1291425158","https://openalex.org/I4210128969"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Nagarajan, Vaishnavh","raw_affiliation_strings":["Equal contribution","Google Research , US"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Equal contribution","institution_ids":[]},{"raw_affiliation_string":"Google Research , US","institution_ids":["https://openalex.org/I1291425158"]}]}],"institutions":[],"countries_distinct_count":2,"institutions_distinct_count":2,"corresponding_author_ids":[],"corresponding_institution_ids":[],"apc_list":null,"apc_paid":null,"fwci":null,"has_fulltext":true,"cited_by_count":6,"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":0.7163000106811523,"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":0.7163000106811523,"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":0.22120000422000885,"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":0.006500000134110451,"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/next-token-prediction","display_name":"next-token prediction","score":0.9990000128746033},{"id":"https://openalex.org/keywords/teacher-forcing","display_name":"teacher forcing","score":0.9039999842643738},{"id":"https://openalex.org/keywords/autoregressive-inference","display_name":"autoregressive inference","score":0.9350000023841858},{"id":"https://openalex.org/keywords/multi-token-prediction","display_name":"multi-token prediction","score":0.6740000247955322},{"id":"https://openalex.org/keywords/transformer","display_name":"transformer models","score":0.7269999980926514},{"id":"https://openalex.org/keywords/mamba","display_name":"Mamba","score":0.9700000286102295},{"id":"https://openalex.org/keywords/planning-tasks","display_name":"planning tasks","score":0.7549999952316284},{"id":"https://openalex.org/keywords/large-language-models","display_name":"large language models","score":0.5540000200271606}],"concepts":[{"id":"https://openalex.org/C48145219","wikidata":"https://www.wikidata.org/wiki/Q1335365","display_name":"Security token","level":2,"score":0.7068099975585938},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.5362452864646912},{"id":"https://openalex.org/C38652104","wikidata":"https://www.wikidata.org/wiki/Q3510521","display_name":"Computer security","level":1,"score":0.16844522953033447}],"mesh":[],"locations_count":2,"locations":[{"id":"pmh:oai:arXiv.org:2403.06963","doi":null,"is_oa":true,"landing_page_url":"http://arxiv.org/abs/2403.06963","pdf_url":"https://arxiv.org/pdf/2403.06963","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.2403.06963","doi":"https://doi.org/10.48550/arxiv.2403.06963","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2403.06963","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:2403.06963","doi":null,"is_oa":true,"landing_page_url":"http://arxiv.org/abs/2403.06963","pdf_url":"https://arxiv.org/pdf/2403.06963","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/W4392737513.pdf","grobid_xml":"https://content.openalex.org/works/W4392737513.grobid-xml"},"referenced_works_count":0,"referenced_works":[],"related_works":["https://openalex.org/W7148186024","https://openalex.org/W4417538281","https://openalex.org/W4407759428","https://openalex.org/W4410000156","https://openalex.org/W7090875166","https://openalex.org/W4404782568","https://openalex.org/W4312121705","https://openalex.org/W4404306241","https://openalex.org/W7138145480","https://openalex.org/W7162423120"],"abstract_inverted_index":{"Can":[0],"a":[1,22,27,74,101,111,151],"mere":[2],"next-token":[3,38,69,93,176],"predictor":[4,94],"faithfully":[5],"model":[6],"human":[7],"intelligence?":[8],"We":[9,99,163,179],"crystallize":[10],"this":[11,143,165],"emerging":[12],"concern":[13],"and":[14,20,43,109,119,171],"correct":[15],"popular":[16,52],"misconceptions":[17],"surrounding":[18],"it,":[19],"advocate":[21],"simple":[23,152],"multi-token":[24],"objective.":[25],"As":[26],"starting":[28],"point,":[29],"we":[30,78,138],"argue":[31],"that":[32,54,63,126,142,157],"the":[33,96,117,120,131,175],"two":[34],"often-conflated":[35],"phases":[36],"of":[37,83,104],"prediction":[39,177],"--":[40,46,128],"autoregressive":[41,59],"inference":[42],"teacher-forced":[44],"training":[45],"must":[47],"be":[48,146],"treated":[49],"distinctly.":[50],"The":[51],"criticism":[53],"errors":[55],"can":[56,86,107,145,167],"compound":[57],"during":[58],"inference,":[60],"crucially":[61],"assumes":[62],"teacher-forcing":[64,85,106],"has":[65],"learned":[66],"an":[67,91],"accurate":[68,92],"predictor.":[70],"This":[71],"assumption":[72],"sidesteps":[73],"more":[75],"deep-rooted":[76],"problem":[77],"expose:":[79],"in":[80,95,125,161],"certain":[81],"classes":[82],"tasks,":[84],"simply":[87],"fail":[88,124],"to":[89,135],"learn":[90],"first":[97],"place.":[98],"describe":[100],"general":[102],"mechanism":[103],"how":[105],"fail,":[108],"design":[110],"minimal":[112],"planning":[113],"task":[114,132],"where":[115],"both":[116],"Transformer":[118],"Mamba":[121],"architecture":[122],"empirically":[123],"manner":[127],"remarkably,":[129],"despite":[130],"being":[133],"straightforward":[134],"learn.":[136],"Finally,":[137],"provide":[139],"preliminary":[140],"evidence":[141],"failure":[144],"resolved":[147],"using":[148,154],"_teacherless_":[149],"training,":[150],"modification":[153],"dummy":[155],"tokens":[156,160],"predicts":[158],"multiple":[159],"advance.":[162],"hope":[164],"finding":[166],"ground":[168],"future":[169],"debates":[170],"inspire":[172],"explorations":[173],"beyond":[174],"paradigm.":[178],"make":[180],"our":[181],"code":[182],"available":[183],"under":[184],"https://github.com/gregorbachmann/Next-Token-Failures":[185]},"counts_by_year":[{"year":2025,"cited_by_count":4},{"year":2024,"cited_by_count":2}],"updated_date":"2026-10-16T09:59:22.235173","created_date":"2024-03-13T00:00:00"}