{"id":"https://openalex.org/W4403580095","doi":"https://doi.org/10.48550/arxiv.2410.13787","title":"Looking Inward: Language Models Can Learn About Themselves by Introspection","display_name":"Looking Inward: Language Models Can Learn About Themselves by Introspection","publication_year":2024,"publication_date":"2024-10-17","ids":{"openalex":"https://openalex.org/W4403580095","doi":"https://doi.org/10.48550/arxiv.2410.13787"},"language":"en","primary_location":{"id":"pmh:oai:arXiv.org:2410.13787","doi":null,"is_oa":true,"landing_page_url":"http://arxiv.org/abs/2410.13787","pdf_url":"https://arxiv.org/pdf/2410.13787","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/2410.13787","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5052496533","display_name":"Felix J. Binder","orcid":"https://orcid.org/0000-0001-5081-7754","observed_orcids":["https://orcid.org/0000-0001-5081-7754"]},"institutions":[{"id":"https://openalex.org/I36258959","display_name":"University of California San Diego","ror":"https://ror.org/0168r3w48","country_code":"US","type":"education","lineage":["https://openalex.org/I36258959"]},{"id":"https://openalex.org/I97018004","display_name":"Stanford University","ror":"https://ror.org/00f54p054","country_code":"US","type":"education","lineage":["https://openalex.org/I97018004"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Binder, Felix J","raw_affiliation_strings":["UC San Diego Stanford University"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"UC San Diego Stanford University","institution_ids":["https://openalex.org/I36258959","https://openalex.org/I97018004"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5050685525","display_name":"James Chua","orcid":null,"observed_orcids":[]},"institutions":[{"id":"https://openalex.org/I36258959","display_name":"University of California San Diego","ror":"https://ror.org/0168r3w48","country_code":"US","type":"education","lineage":["https://openalex.org/I36258959"]},{"id":"https://openalex.org/I97018004","display_name":"Stanford University","ror":"https://ror.org/00f54p054","country_code":"US","type":"education","lineage":["https://openalex.org/I97018004"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Chua, James","raw_affiliation_strings":["UC San Diego Stanford University"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"UC San Diego Stanford University","institution_ids":["https://openalex.org/I36258959","https://openalex.org/I97018004"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5114338125","display_name":"Tomek Korbak","orcid":null,"observed_orcids":[]},"institutions":[{"id":"https://openalex.org/I57206974","display_name":"New York University","ror":"https://ror.org/0190ak572","country_code":"US","type":"education","lineage":["https://openalex.org/I57206974"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Korbak, Tomek","raw_affiliation_strings":["New York University"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"New York University","institution_ids":["https://openalex.org/I57206974"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5094579744","display_name":"Henry Sleight","orcid":null,"observed_orcids":[]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Sleight, Henry","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5102452417","display_name":"John Hughes","orcid":null,"observed_orcids":[]},"institutions":[{"id":"https://openalex.org/I95457486","display_name":"University of California, Berkeley","ror":"https://ror.org/01an7q238","country_code":"US","type":"education","lineage":["https://openalex.org/I95457486"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Hughes, John","raw_affiliation_strings":["UC Berkeley"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"UC Berkeley","institution_ids":["https://openalex.org/I95457486"]}]},{"author_position":"middle","author":{"id":null,"display_name":"Long, Robert","orcid":null,"observed_orcids":[]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Long, Robert","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5091112967","display_name":"Ethan Perez","orcid":null,"observed_orcids":[]},"institutions":[{"id":"https://openalex.org/I36258959","display_name":"University of California San Diego","ror":"https://ror.org/0168r3w48","country_code":"US","type":"education","lineage":["https://openalex.org/I36258959"]},{"id":"https://openalex.org/I97018004","display_name":"Stanford University","ror":"https://ror.org/00f54p054","country_code":"US","type":"education","lineage":["https://openalex.org/I97018004"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Perez, Ethan","raw_affiliation_strings":["UC San Diego Stanford University"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"UC San Diego Stanford University","institution_ids":["https://openalex.org/I36258959","https://openalex.org/I97018004"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5113139169","display_name":"Miles Turpin","orcid":null,"observed_orcids":[]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Turpin, Miles","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5089789418","display_name":"Owain Evans","orcid":null,"observed_orcids":[]},"institutions":[{"id":"https://openalex.org/I95457486","display_name":"University of California, Berkeley","ror":"https://ror.org/01an7q238","country_code":"US","type":"education","lineage":["https://openalex.org/I95457486"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Evans, Owain","raw_affiliation_strings":["UC Berkeley"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"UC Berkeley","institution_ids":["https://openalex.org/I95457486"]}]}],"institutions":[],"countries_distinct_count":1,"institutions_distinct_count":4,"corresponding_author_ids":[],"corresponding_institution_ids":[],"apc_list":null,"apc_paid":null,"fwci":null,"has_fulltext":true,"cited_by_count":108,"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/T12026","display_name":"Explainable Artificial Intelligence (XAI)","score":0.32010000944137573,"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/T12026","display_name":"Explainable Artificial Intelligence (XAI)","score":0.32010000944137573,"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/T10258","display_name":"Philosophy and Theoretical Science","score":0.23170000314712524,"subfield":{"id":"https://openalex.org/subfields/1211","display_name":"Philosophy"},"field":{"id":"https://openalex.org/fields/12","display_name":"Arts and Humanities"},"domain":{"id":"https://openalex.org/domains/2","display_name":"Social Sciences"}},{"id":"https://openalex.org/T10028","display_name":"Topic Modeling","score":0.1784999966621399,"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/self-reflection","display_name":"self-reflection","score":0.8550000190734863},{"id":"https://openalex.org/keywords/large-language-models","display_name":"large language models","score":0.9049999713897705},{"id":"https://openalex.org/keywords/self-prediction","display_name":"self-prediction","score":0.8730000257492065},{"id":"https://openalex.org/keywords/model-interpretability","display_name":"model interpretability","score":0.703000009059906},{"id":"https://openalex.org/keywords/fine-tuning","display_name":"fine-tuning","score":0.6389999985694885},{"id":"https://openalex.org/keywords/gpt-4","display_name":"GPT-4","score":0.7829999923706055},{"id":"https://openalex.org/keywords/llama-3","display_name":"Llama 3","score":0.9089999794960022},{"id":"https://openalex.org/keywords/gpt-4o","display_name":"GPT-4o","score":0.703000009059906},{"id":"https://openalex.org/keywords/out-of-distribution-generalization","display_name":"out-of-distribution generalization","score":0.7289999723434448}],"concepts":[{"id":"https://openalex.org/C129671850","wikidata":"https://www.wikidata.org/wiki/Q210501","display_name":"Introspection","level":2,"score":0.9466903209686279},{"id":"https://openalex.org/C41895202","wikidata":"https://www.wikidata.org/wiki/Q8162","display_name":"Linguistics","level":1,"score":0.43610602617263794},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.39960479736328125},{"id":"https://openalex.org/C15744967","wikidata":"https://www.wikidata.org/wiki/Q9418","display_name":"Psychology","level":0,"score":0.38629603385925293},{"id":"https://openalex.org/C180747234","wikidata":"https://www.wikidata.org/wiki/Q23373","display_name":"Cognitive psychology","level":1,"score":0.3284924030303955},{"id":"https://openalex.org/C138885662","wikidata":"https://www.wikidata.org/wiki/Q5891","display_name":"Philosophy","level":0,"score":0.1591426134109497}],"mesh":[],"locations_count":2,"locations":[{"id":"pmh:oai:arXiv.org:2410.13787","doi":null,"is_oa":true,"landing_page_url":"http://arxiv.org/abs/2410.13787","pdf_url":"https://arxiv.org/pdf/2410.13787","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.2410.13787","doi":"https://doi.org/10.48550/arxiv.2410.13787","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2410.13787","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:2410.13787","doi":null,"is_oa":true,"landing_page_url":"http://arxiv.org/abs/2410.13787","pdf_url":"https://arxiv.org/pdf/2410.13787","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":[{"id":"https://openalex.org/F4320315389","display_name":"Open Philanthropy Project","ror":"https://ror.org/004d1k391"}],"has_content":{"pdf":true,"grobid_xml":true},"content_urls":{"pdf":"https://content.openalex.org/works/W4403580095.pdf","grobid_xml":"https://content.openalex.org/works/W4403580095.grobid-xml"},"referenced_works_count":29,"referenced_works":["https://openalex.org/W2794325560","https://openalex.org/W3168867926","https://openalex.org/W3202712981","https://openalex.org/W3205287819","https://openalex.org/W3215698779","https://openalex.org/W4281748205","https://openalex.org/W4285429195","https://openalex.org/W4294435850","https://openalex.org/W4378509375","https://openalex.org/W4385327568","https://openalex.org/W4386494224","https://openalex.org/W4387892136","https://openalex.org/W4388747740","https://openalex.org/W4388786517","https://openalex.org/W4389768135","https://openalex.org/W4389911544","https://openalex.org/W4391272510","https://openalex.org/W4391835415","https://openalex.org/W4392678667","https://openalex.org/W4393336283","https://openalex.org/W4396945540","https://openalex.org/W4399694448","https://openalex.org/W4402671982","https://openalex.org/W4410357125","https://openalex.org/W4415796979","https://openalex.org/W4415797148","https://openalex.org/W4415797398","https://openalex.org/W4415798308","https://openalex.org/W4415800371"],"related_works":["https://openalex.org/W7140186876","https://openalex.org/W4415100619","https://openalex.org/W4417078358","https://openalex.org/W4417448492","https://openalex.org/W7162561219","https://openalex.org/W7118379034","https://openalex.org/W4414990387","https://openalex.org/W7110478437","https://openalex.org/W4387404506","https://openalex.org/W4413127078"],"abstract_inverted_index":{"Humans":[0],"acquire":[1],"knowledge":[2,43],"by":[3,10,127,135],"observing":[4],"the":[5,79,114,128,151,158,234],"external":[6,33],"world,":[7],"but":[8,54],"also":[9],"introspection.":[11,245],"Introspection":[12],"gives":[13],"a":[14,61,71,164,172],"person":[15],"privileged":[16,195],"access":[17,196],"to":[18,32,138,197,206,228,249],"their":[19,142],"current":[20],"state":[21],"of":[22,68,117,141],"mind":[23],"(e.g.,":[24],"thoughts":[25],"and":[26,86,108,202,223],"feelings)":[27],"that":[28,44,192,233],"is":[29,45,183,191,215],"not":[30,46,123],"accessible":[31],"observers.":[34],"Can":[35],"LLMs":[36,137],"introspect?":[37],"We":[38,132],"define":[39],"introspection":[40,134,267],"as":[41,103],"acquiring":[42],"contained":[47],"in":[48,145,176,239],"or":[49,106,160,278],"derived":[50],"from":[51,57],"training":[52,130],"data":[53],"instead":[55],"originates":[56],"internal":[58,73,100],"states.":[59,119],"Such":[60,120],"capability":[62],"could":[63,76,110],"enhance":[64],"model":[65,80,92,165,174,235],"interpretability.":[66],"Instead":[67],"painstakingly":[69],"analyzing":[70],"model's":[72,129],"workings,":[74],"we":[75,231,256,264,271],"simply":[77],"ask":[78],"about":[81,113],"its":[82,198,251,259],"beliefs,":[83],"world":[84],"models,":[85],"goals.":[87],"More":[88],"speculatively,":[89],"an":[90],"introspective":[91],"might":[93],"self-report":[94],"on":[95,185,268,274],"whether":[96],"it":[97,169,205],"possesses":[98],"certain":[99],"states":[101],"such":[102],"subjective":[104],"feelings":[105],"desires":[107],"this":[109,203],"inform":[111],"us":[112],"moral":[115],"status":[116],"these":[118],"self-reports":[121],"would":[122,154],"be":[124],"entirely":[125],"dictated":[126],"data.":[131],"study":[133],"finetuning":[136],"predict":[139,207,229,250],"properties":[140],"own":[143,199],"behavior":[144,179,252],"hypothetical":[146],"scenarios.":[147],"For":[148],"example,":[149],"\"Given":[150],"input":[152],"P,":[153],"your":[155],"output":[156],"favor":[157],"short-":[159],"long-term":[161],"option?\"":[162],"If":[163],"M1":[166,193,236,247],"can":[167],"introspect,":[168],"should":[170],"outperform":[171],"different":[173],"M2":[175,182,211,214,238],"predicting":[177,240],"M1's":[178,186],"even":[180,254],"if":[181,213],"trained":[184],"ground-truth":[187,260],"behavior.":[188,261],"The":[189],"idea":[190],"has":[194],"behavioral":[200],"tendencies,":[201],"enables":[204],"itself":[208],"better":[209],"than":[210],"(even":[212],"generally":[216],"stronger).":[217],"In":[218],"experiments":[219],"with":[220],"GPT-4,":[221],"GPT-4o,":[222],"Llama-3":[224],"models":[225],"(each":[226],"finetuned":[227],"itself),":[230],"find":[232],"outperforms":[237],"itself,":[241],"providing":[242],"evidence":[243],"for":[244],"Notably,":[246],"continues":[248],"accurately":[253],"after":[255],"intentionally":[257],"modify":[258],"However,":[262],"while":[263],"successfully":[265],"elicit":[266],"simple":[269],"tasks,":[270],"are":[272],"unsuccessful":[273],"more":[275],"complex":[276],"tasks":[277],"those":[279],"requiring":[280],"out-of-distribution":[281],"generalization.":[282]},"counts_by_year":[{"year":2026,"cited_by_count":71},{"year":2025,"cited_by_count":33},{"year":2024,"cited_by_count":3},{"year":2022,"cited_by_count":1}],"updated_date":"2026-11-07T11:25:07.637180","created_date":"2024-10-21T00:00:00"}
