{"id":"https://openalex.org/W4415797451","doi":"https://doi.org/10.52202/079017-1939","title":"How Do Large Language Models Acquire Factual Knowledge During Pretraining?","display_name":"How Do Large Language Models Acquire Factual Knowledge During Pretraining?","publication_year":2024,"publication_date":"2024-01-01","ids":{"openalex":"https://openalex.org/W4415797451","doi":"https://doi.org/10.52202/079017-1939"},"language":"en","primary_location":{"id":"doi:10.52202/079017-1939","doi":"https://doi.org/10.52202/079017-1939","is_oa":false,"landing_page_url":"https://doi.org/10.52202/079017-1939","pdf_url":null,"source":{"id":"https://openalex.org/S4363606243","display_name":"neural information processing systems","issn_l":"1049-5258","issn":["1049-5258"],"is_oa":false,"is_in_doaj":false,"is_core":true,"listed_in":["cwts-core","norway-2"],"host_organization":null,"host_organization_name":null,"host_organization_lineage":[],"host_organization_lineage_names":[],"type":"conference"},"license":null,"license_id":null,"version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Advances in Neural Information Processing Systems 37","raw_type":"proceedings-article"},"volume":null,"type":"conference-paper","is_traditional":true,"indexed_in":["arxiv","crossref","datacite"],"open_access":{"is_oa":true,"oa_status":"green","oa_url":"https://arxiv.org/pdf/2406.11813","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5066359770","display_name":"Du-Seong Chang","orcid":null,"observed_orcids":[]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Hoyeon Chang","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":null,"display_name":"Jinho Park","orcid":null,"observed_orcids":[]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Jinho Park","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":null,"display_name":"Seonghyeon Ye","orcid":null,"observed_orcids":[]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Seonghyeon Ye","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":null,"display_name":"Sohee Yang","orcid":null,"observed_orcids":[]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Sohee Yang","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5049262418","display_name":"Youngkyung Seo","orcid":null,"observed_orcids":[]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Youngkyung Seo","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5066359770","display_name":"Du-Seong Chang","orcid":null,"observed_orcids":[]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Du-Seong Chang","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5087565126","display_name":"Minjoon Seo","orcid":null,"observed_orcids":[]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Minjoon Seo","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":9.6195,"has_fulltext":true,"cited_by_count":63,"citation_normalized_percentile":{"value":0.98592722,"is_in_top_1_percent":false,"is_in_top_10_percent":true},"cited_by_percentile_year":{"min":99,"max":100},"biblio":{"volume":null,"issue":null,"first_page":"60626","last_page":"60668"},"is_retracted":false,"is_paratext":false,"is_xpac":false,"primary_topic":{"id":"https://openalex.org/T10028","display_name":"Topic Modeling","score":0.9717000126838684,"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":0.9717000126838684,"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":0.010599999688565731,"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":0.0071000000461936,"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":0.9480000138282776},{"id":"https://openalex.org/keywords/pretraining","display_name":"pre-training","score":0.8790000081062317},{"id":"https://openalex.org/keywords/knowledge-forgetting","display_name":"knowledge forgetting","score":0.640999972820282},{"id":"https://openalex.org/keywords/memorization","display_name":"memorization","score":0.9150000214576721},{"id":"https://openalex.org/keywords/generalization","display_name":"generalization","score":0.6140000224113464},{"id":"https://openalex.org/keywords/batch-size","display_name":"batch size","score":0.8149999976158142},{"id":"https://openalex.org/keywords/power-law-relationship","display_name":"power-law relationship","score":0.6869999766349792},{"id":"https://openalex.org/keywords/long-tail-knowledge","display_name":"long-tail knowledge","score":0.8809999823570251}],"concepts":[{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.5972999930381775},{"id":"https://openalex.org/C204321447","wikidata":"https://www.wikidata.org/wiki/Q30642","display_name":"Natural language processing","level":1,"score":0.43849998712539673},{"id":"https://openalex.org/C41895202","wikidata":"https://www.wikidata.org/wiki/Q8162","display_name":"Linguistics","level":1,"score":0.4058000147342682},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.4036000072956085},{"id":"https://openalex.org/C137293760","wikidata":"https://www.wikidata.org/wiki/Q3621696","display_name":"Language model","level":2,"score":0.33410000801086426},{"id":"https://openalex.org/C195324797","wikidata":"https://www.wikidata.org/wiki/Q33742","display_name":"Natural language","level":2,"score":0.31869998574256897},{"id":"https://openalex.org/C2983448237","wikidata":"https://www.wikidata.org/wiki/Q1078276","display_name":"Language understanding","level":2,"score":0.2971000075340271},{"id":"https://openalex.org/C2776401178","wikidata":"https://www.wikidata.org/wiki/Q12050496","display_name":"Feature (linguistics)","level":2,"score":0.2784000039100647},{"id":"https://openalex.org/C179603123","wikidata":"https://www.wikidata.org/wiki/Q1941921","display_name":"Modeling language","level":3,"score":0.2750000059604645},{"id":"https://openalex.org/C184337299","wikidata":"https://www.wikidata.org/wiki/Q1437428","display_name":"Semantics (computer science)","level":2,"score":0.2671999931335449},{"id":"https://openalex.org/C161301231","wikidata":"https://www.wikidata.org/wiki/Q3478658","display_name":"Knowledge representation and reasoning","level":2,"score":0.2653999924659729},{"id":"https://openalex.org/C15744967","wikidata":"https://www.wikidata.org/wiki/Q9418","display_name":"Psychology","level":0,"score":0.2615000009536743},{"id":"https://openalex.org/C129792486","wikidata":"https://www.wikidata.org/wiki/Q1050419","display_name":"Language identification","level":3,"score":0.2572999894618988},{"id":"https://openalex.org/C188147891","wikidata":"https://www.wikidata.org/wiki/Q147638","display_name":"Cognitive science","level":1,"score":0.2540999948978424},{"id":"https://openalex.org/C2780791683","wikidata":"https://www.wikidata.org/wiki/Q846785","display_name":"Action (physics)","level":2,"score":0.25360000133514404}],"mesh":[],"locations_count":3,"locations":[{"id":"doi:10.52202/079017-1939","doi":"https://doi.org/10.52202/079017-1939","is_oa":false,"landing_page_url":"https://doi.org/10.52202/079017-1939","pdf_url":null,"source":{"id":"https://openalex.org/S4363606243","display_name":"neural information processing systems","issn_l":"1049-5258","issn":["1049-5258"],"is_oa":false,"is_in_doaj":false,"is_core":true,"listed_in":["cwts-core","norway-2"],"host_organization":null,"host_organization_name":null,"host_organization_lineage":[],"host_organization_lineage_names":[],"type":"conference"},"license":null,"license_id":null,"version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Advances in Neural Information Processing Systems 37","raw_type":"proceedings-article"},{"id":"pmh:oai:arXiv.org:2406.11813","doi":null,"is_oa":true,"landing_page_url":"http://arxiv.org/abs/2406.11813","pdf_url":"https://arxiv.org/pdf/2406.11813","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.2406.11813","doi":"https://doi.org/10.48550/arxiv.2406.11813","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2406.11813","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:2406.11813","doi":null,"is_oa":true,"landing_page_url":"http://arxiv.org/abs/2406.11813","pdf_url":"https://arxiv.org/pdf/2406.11813","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":[{"id":"https://openalex.org/G3736587842","display_name":null,"funder_award_id":"RS-2019-II190075","funder_id":"https://openalex.org/F4320335489","funder_display_name":"Institute for Information and Communications Technology Promotion"},{"id":"https://openalex.org/G4468082379","display_name":null,"funder_award_id":"RS-2019-II190075","funder_id":"https://openalex.org/F4320328359","funder_display_name":"Ministry of Science and ICT, South Korea"}],"funders":[{"id":"https://openalex.org/F4320324161","display_name":"Korea Advanced Institute of Science and Technology","ror":"https://ror.org/05apxxy63"},{"id":"https://openalex.org/F4320328359","display_name":"Ministry of Science and ICT, South Korea","ror":"https://ror.org/01wpjm123"},{"id":"https://openalex.org/F4320335489","display_name":"Institute for Information and Communications Technology Promotion","ror":"https://ror.org/01g0hqq23"}],"has_content":{"grobid_xml":true,"pdf":true},"content_urls":{"pdf":"https://content.openalex.org/works/W4415797451.pdf","grobid_xml":"https://content.openalex.org/works/W4415797451.grobid-xml"},"referenced_works_count":8,"referenced_works":["https://openalex.org/W3117576675","https://openalex.org/W3152884768","https://openalex.org/W3213280751","https://openalex.org/W4322718191","https://openalex.org/W4327810158","https://openalex.org/W4389520380","https://openalex.org/W4402671286","https://openalex.org/W4404782801"],"related_works":["https://openalex.org/W4415063744","https://openalex.org/W4410356661","https://openalex.org/W4415326889","https://openalex.org/W4415981209","https://openalex.org/W4416034846","https://openalex.org/W7208725022","https://openalex.org/W7153135527","https://openalex.org/W4404792828","https://openalex.org/W4406072499","https://openalex.org/W6929243244"],"abstract_inverted_index":{"Despite":[0],"the":[1,20,51,73,119,139,146,180,189,193],"recent":[2],"observation":[3],"that":[4,63,128,166],"large":[5],"language":[6],"models":[7],"(LLMs)":[8],"can":[9,117,168],"store":[10],"substantial":[11],"factual":[12,26,40,54,80,98,129,142],"knowledge,":[13,99],"there":[14,83],"is":[15,84,155],"a":[16,85],"limited":[17],"understanding":[18],"of":[19,22,53,93,97,141,176,183,191],"mechanisms":[21],"how":[23,37],"they":[24],"acquire":[25,39,77],"knowledge":[27,41,55,130,143,187],"through":[28],"pretraining.":[29,43,58],"This":[30],"work":[31],"addresses":[32],"this":[33,153,162],"gap":[34],"by":[35,136,157],"studying":[36],"LLMs":[38,101,112,184],"during":[42,57],"The":[44],"findings":[45],"reveal":[46],"several":[47],"important":[48],"insights":[49],"into":[50],"dynamics":[52],"acquisition":[56,131],"First,":[59],"counterintuitively,":[60],"we":[61,164,167],"observe":[62],"pretraining":[64,134,147,194],"on":[65,161,185],"more":[66],"data":[67,106,148],"shows":[68],"no":[69],"significant":[70],"improvement":[71],"in":[72,132,145],"model's":[74],"capability":[75],"to":[76,122],"and":[78,91,95,100,188],"maintain":[79],"knowledge.":[81],"Next,":[82],"power-law":[86],"relationship":[87],"between":[88],"training":[89,105,111],"steps":[90],"forgetting":[92],"memorization":[94],"generalization":[96],"trained":[102],"with":[103,113],"duplicated":[104],"exhibit":[107],"faster":[108],"forgetting.":[109,123,159],"Third,":[110],"larger":[114],"batch":[115],"sizes":[116],"enhance":[118],"models'":[120],"robustness":[121],"Overall,":[124],"our":[125],"observations":[126],"suggest":[127],"LLM":[133],"occurs":[135],"progressively":[137],"increasing":[138],"probability":[140],"presented":[144],"at":[149],"each":[150],"step.":[151],"However,":[152],"increase":[154],"diluted":[156],"subsequent":[158],"Based":[160],"interpretation,":[163],"demonstrate":[165],"provide":[169],"plausible":[170],"explanations":[171],"for":[172],"recently":[173],"observed":[174],"behaviors":[175],"LLMs,":[177],"such":[178],"as":[179],"poor":[181],"performance":[182],"long-tail":[186],"benefits":[190],"deduplicating":[192],"corpus.":[195]},"counts_by_year":[{"year":2026,"cited_by_count":27},{"year":2025,"cited_by_count":24},{"year":2024,"cited_by_count":12}],"updated_date":"2026-11-10T11:39:43.021295","created_date":"2025-11-03T00:00:00"}