{"id":"https://openalex.org/W4389518334","doi":"https://doi.org/10.18653/v1/2023.blackboxnlp-1.22","title":"Layered Bias: Interpreting Bias in Pretrained Large Language Models","display_name":"Layered Bias: Interpreting Bias in Pretrained Large Language Models","publication_year":2023,"publication_date":"2023-01-01","ids":{"openalex":"https://openalex.org/W4389518334","doi":"https://doi.org/10.18653/v1/2023.blackboxnlp-1.22"},"language":"en","primary_location":{"id":"doi:10.18653/v1/2023.blackboxnlp-1.22","doi":"https://doi.org/10.18653/v1/2023.blackboxnlp-1.22","is_oa":true,"landing_page_url":"https://doi.org/10.18653/v1/2023.blackboxnlp-1.22","pdf_url":"https://aclanthology.org/2023.blackboxnlp-1.22.pdf","source":{"id":"https://openalex.org/S7407088969","display_name":"BlackboxNLP Workshop on Analyzing and Interpreting Neural Networks for NLP (BlackboxNLP)","issn_l":null,"issn":null,"is_oa":false,"is_in_doaj":false,"is_core":false,"listed_in":[],"host_organization":null,"host_organization_name":null,"host_organization_lineage":[],"host_organization_lineage_names":[],"type":"conference"},"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Proceedings of the 6th BlackboxNLP Workshop: Analyzing and Interpreting Neural Networks for NLP","raw_type":"proceedings-article"},"volume":null,"type":"conference-paper","is_traditional":true,"indexed_in":["crossref"],"open_access":{"is_oa":true,"oa_status":"hybrid","oa_url":"https://aclanthology.org/2023.blackboxnlp-1.22.pdf","any_repository_has_fulltext":false},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5056629800","display_name":"Nirmalendu Prakash","orcid":"https://orcid.org/0009-0009-3144-9760","observed_orcids":["https://orcid.org/0009-0009-3144-9760"]},"institutions":[{"id":"https://openalex.org/I152815399","display_name":"Singapore University of Technology and Design","ror":"https://ror.org/05j6fvn87","country_code":"SG","type":"education","lineage":["https://openalex.org/I152815399"]}],"countries":["SG"],"is_corresponding":true,"raw_author_name":"Nirmalendu Prakash","raw_affiliation_strings":["Singapore University of Technology and Design"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Singapore University of Technology and Design","institution_ids":["https://openalex.org/I152815399"]}]},{"author_position":"last","author":{"id":"https://openalex.org/A5089793938","display_name":"Roy Ka-Wei Lee","orcid":"https://orcid.org/0000-0002-1986-7750","observed_orcids":["https://orcid.org/0000-0002-1986-7750"]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Roy Ka-Wei Lee","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]}],"institutions":[],"countries_distinct_count":1,"institutions_distinct_count":1,"corresponding_author_ids":["https://openalex.org/A5056629800"],"corresponding_institution_ids":["https://openalex.org/I152815399"],"apc_list":null,"apc_paid":null,"fwci":1.6571,"has_fulltext":true,"cited_by_count":15,"citation_normalized_percentile":{"value":0.88355399,"is_in_top_1_percent":false,"is_in_top_10_percent":false},"cited_by_percentile_year":{"min":95,"max":99},"biblio":{"volume":null,"issue":null,"first_page":"284","last_page":"295"},"is_retracted":false,"is_paratext":false,"is_xpac":false,"primary_topic":{"id":"https://openalex.org/T10883","display_name":"Ethics and Social Impacts of AI","score":0.86080002784729,"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/T10883","display_name":"Ethics and Social Impacts of AI","score":0.86080002784729,"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/T10028","display_name":"Topic Modeling","score":0.05180000141263008,"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/T12026","display_name":"Explainable Artificial Intelligence (XAI)","score":0.030700000002980232,"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":0.6840000152587891},{"id":"https://openalex.org/keywords/social-bias","display_name":"social bias","score":0.8059999942779541},{"id":"https://openalex.org/keywords/bias-mitigation","display_name":"bias mitigation","score":0.890999972820282},{"id":"https://openalex.org/keywords/counterfactual-data-augmentation","display_name":"counterfactual data augmentation","score":0.9800000190734863},{"id":"https://openalex.org/keywords/logit-lens","display_name":"logit lens","score":0.8270000219345093},{"id":"https://openalex.org/keywords/transformer-layers","display_name":"transformer layers","score":0.8050000071525574},{"id":"https://openalex.org/keywords/opt","display_name":"OPT","score":0.5590000152587891},{"id":"https://openalex.org/keywords/llama","display_name":"llamas","score":0.9959999918937683},{"id":"https://openalex.org/keywords/llama-2","display_name":"Llama 2","score":0.9210000038146973},{"id":"https://openalex.org/keywords/bias-assessment","display_name":"bias assessment","score":0.6579999923706055}],"concepts":[{"id":"https://openalex.org/C2779458634","wikidata":"https://www.wikidata.org/wiki/Q24963715","display_name":"Debiasing","level":2,"score":0.9757161140441895},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.7162086963653564},{"id":"https://openalex.org/C108650721","wikidata":"https://www.wikidata.org/wiki/Q1783253","display_name":"Counterfactual thinking","level":2,"score":0.6724812984466553},{"id":"https://openalex.org/C140331021","wikidata":"https://www.wikidata.org/wiki/Q1868104","display_name":"Logit","level":2,"score":0.5125920176506042},{"id":"https://openalex.org/C185798385","wikidata":"https://www.wikidata.org/wiki/Q1161707","display_name":"Benchmark (surveying)","level":2,"score":0.5001966953277588},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.48752090334892273},{"id":"https://openalex.org/C119857082","wikidata":"https://www.wikidata.org/wiki/Q2539","display_name":"Machine learning","level":1,"score":0.4836631715297699},{"id":"https://openalex.org/C165064840","wikidata":"https://www.wikidata.org/wiki/Q1321061","display_name":"Matching (statistics)","level":2,"score":0.4783441722393036},{"id":"https://openalex.org/C203005215","wikidata":"https://www.wikidata.org/wiki/Q79798","display_name":"Machine translation","level":2,"score":0.4678899347782135},{"id":"https://openalex.org/C204321447","wikidata":"https://www.wikidata.org/wiki/Q30642","display_name":"Natural language processing","level":1,"score":0.42986732721328735},{"id":"https://openalex.org/C50644808","wikidata":"https://www.wikidata.org/wiki/Q192776","display_name":"Artificial neural network","level":2,"score":0.42246368527412415},{"id":"https://openalex.org/C15744967","wikidata":"https://www.wikidata.org/wiki/Q9418","display_name":"Psychology","level":0,"score":0.15803101658821106},{"id":"https://openalex.org/C105795698","wikidata":"https://www.wikidata.org/wiki/Q12483","display_name":"Statistics","level":1,"score":0.1093091368675232},{"id":"https://openalex.org/C77805123","wikidata":"https://www.wikidata.org/wiki/Q161272","display_name":"Social psychology","level":1,"score":0.0},{"id":"https://openalex.org/C33923547","wikidata":"https://www.wikidata.org/wiki/Q395","display_name":"Mathematics","level":0,"score":0.0},{"id":"https://openalex.org/C13280743","wikidata":"https://www.wikidata.org/wiki/Q131089","display_name":"Geodesy","level":1,"score":0.0},{"id":"https://openalex.org/C188147891","wikidata":"https://www.wikidata.org/wiki/Q147638","display_name":"Cognitive science","level":1,"score":0.0},{"id":"https://openalex.org/C205649164","wikidata":"https://www.wikidata.org/wiki/Q1071","display_name":"Geography","level":0,"score":0.0}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.18653/v1/2023.blackboxnlp-1.22","doi":"https://doi.org/10.18653/v1/2023.blackboxnlp-1.22","is_oa":true,"landing_page_url":"https://doi.org/10.18653/v1/2023.blackboxnlp-1.22","pdf_url":"https://aclanthology.org/2023.blackboxnlp-1.22.pdf","source":{"id":"https://openalex.org/S7407088969","display_name":"BlackboxNLP Workshop on Analyzing and Interpreting Neural Networks for NLP (BlackboxNLP)","issn_l":null,"issn":null,"is_oa":false,"is_in_doaj":false,"is_core":false,"listed_in":[],"host_organization":null,"host_organization_name":null,"host_organization_lineage":[],"host_organization_lineage_names":[],"type":"conference"},"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Proceedings of the 6th BlackboxNLP Workshop: Analyzing and Interpreting Neural Networks for NLP","raw_type":"proceedings-article"}],"best_oa_location":{"id":"doi:10.18653/v1/2023.blackboxnlp-1.22","doi":"https://doi.org/10.18653/v1/2023.blackboxnlp-1.22","is_oa":true,"landing_page_url":"https://doi.org/10.18653/v1/2023.blackboxnlp-1.22","pdf_url":"https://aclanthology.org/2023.blackboxnlp-1.22.pdf","source":{"id":"https://openalex.org/S7407088969","display_name":"BlackboxNLP Workshop on Analyzing and Interpreting Neural Networks for NLP (BlackboxNLP)","issn_l":null,"issn":null,"is_oa":false,"is_in_doaj":false,"is_core":false,"listed_in":[],"host_organization":null,"host_organization_name":null,"host_organization_lineage":[],"host_organization_lineage_names":[],"type":"conference"},"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Proceedings of the 6th BlackboxNLP Workshop: Analyzing and Interpreting Neural Networks for NLP","raw_type":"proceedings-article"},"sustainable_development_goals":[],"study_designs":[],"awards":[],"funders":[],"has_content":{"pdf":true,"grobid_xml":true},"content_urls":{"pdf":"https://content.openalex.org/works/W4389518334.pdf","grobid_xml":"https://content.openalex.org/works/W4389518334.grobid-xml"},"referenced_works_count":40,"referenced_works":["https://openalex.org/W2516809705","https://openalex.org/W2594633041","https://openalex.org/W2618851150","https://openalex.org/W2626778328","https://openalex.org/W2922862135","https://openalex.org/W2950018712","https://openalex.org/W2950866572","https://openalex.org/W2963526187","https://openalex.org/W2970820321","https://openalex.org/W2971307358","https://openalex.org/W2971904186","https://openalex.org/W2972413484","https://openalex.org/W2984662393","https://openalex.org/W3016682301","https://openalex.org/W3019416653","https://openalex.org/W3030163527","https://openalex.org/W3088599783","https://openalex.org/W3089430725","https://openalex.org/W3092475477","https://openalex.org/W3104142662","https://openalex.org/W3123340107","https://openalex.org/W3134354193","https://openalex.org/W3168584517","https://openalex.org/W3168867926","https://openalex.org/W3172205429","https://openalex.org/W3206487987","https://openalex.org/W3207316473","https://openalex.org/W4224308101","https://openalex.org/W4229005866","https://openalex.org/W4322718191","https://openalex.org/W4327526719","https://openalex.org/W4384811843","https://openalex.org/W4384918448","https://openalex.org/W4385571791","https://openalex.org/W4385572928","https://openalex.org/W4385573981","https://openalex.org/W4385574113","https://openalex.org/W4401043228","https://openalex.org/W4411120468","https://openalex.org/W7133236469"],"related_works":["https://openalex.org/W7127987255","https://openalex.org/W4399837132","https://openalex.org/W4389611182","https://openalex.org/W7129205257","https://openalex.org/W7155576697","https://openalex.org/W4401043228","https://openalex.org/W7124294891","https://openalex.org/W6966777154","https://openalex.org/W4415721138","https://openalex.org/W7126183608"],"abstract_inverted_index":{"Large":[0],"language":[1,13],"models":[2],"(LLMs)":[3],"like":[4,42],"GPT":[5],"and":[6,23,47,86,109,111,125,127],"PALM":[7],"have":[8,30,38],"excelled":[9],"in":[10,55],"numerous":[11],"natural":[12],"processing":[14],"(NLP)":[15],"tasks":[16],"such":[17],"as":[18],"text":[19],"generation,":[20],"question":[21],"answering,":[22],"translation.However,":[24],"they":[25],"are":[26,90,116],"also":[27],"found":[28],"to":[29,92],"inherent":[31],"social":[32],"biases.To":[33],"address":[34],"this,":[35],"recent":[36],"studies":[37],"proposed":[39],"debiasing":[40,84],"techniques":[41],"iterative":[43],"nullspace":[44],"projection":[45],"(INLP)":[46],"Counterfactual":[48],"Data":[49],"Augmentation":[50],"(CDA).Additionally,":[51],"there's":[52],"growing":[53],"interest":[54],"understanding":[56],"the":[57,81,134],"intricacies":[58],"of":[59,83],"these":[60],"models.Some":[61],"researchers":[62],"focus":[63],"on":[64,118],"individual":[65],"neural":[66],"units,":[67],"while":[68],"others":[69],"examine":[70],"specific":[71],"layers.In":[72],"our":[73],"study,":[74],"we":[75,102,128],"benchmark":[76],"newly":[77],"released":[78],"models,":[79],"assess":[80],"impact":[82],"methods,":[85],"investigate":[87],"how":[88],"biases":[89],"linked":[91],"different":[93],"transformer":[94],"layers":[95],"using":[96,133],"a":[97,130],"method":[98],"called":[99],"Logit":[100,135],"Lens.Specifically,":[101],"evaluate":[103],"three":[104],"modern":[105],"LLMs:":[106],"OPT,":[107],"LLaMA,":[108],"LLaMA2,":[110],"their":[112],"debiased":[113],"versions.Our":[114],"experiments":[115],"based":[117],"two":[119],"popular":[120],"bias":[121],"evaluation":[122],"datasets,":[123],"Stere-oSet":[124],"CrowS-Pairs,":[126],"perform":[129],"layerby-layer":[131],"analysis":[132],"Lens.":[136]},"counts_by_year":[{"year":2026,"cited_by_count":4},{"year":2025,"cited_by_count":8},{"year":2024,"cited_by_count":3}],"updated_date":"2026-11-07T11:25:07.637180","created_date":"2023-12-10T00:00:00"}