{"id":"https://openalex.org/W4412944612","doi":"https://doi.org/10.18653/v1/2025.findings-acl.1237","title":"Removing Prompt-template Bias in Reinforcement Learning from Human Feedback","display_name":"Removing Prompt-template Bias in Reinforcement Learning from Human Feedback","publication_year":2025,"publication_date":"2025-01-01","ids":{"openalex":"https://openalex.org/W4412944612","doi":"https://doi.org/10.18653/v1/2025.findings-acl.1237"},"language":"en","primary_location":{"id":"doi:10.18653/v1/2025.findings-acl.1237","is_oa":true,"landing_page_url":"https://doi.org/10.18653/v1/2025.findings-acl.1237","pdf_url":"https://aclanthology.org/2025.findings-acl.1237.pdf","source":null,"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Findings of the Association for Computational Linguistics: ACL 2025","raw_type":"proceedings-article"},"type":"article","indexed_in":["crossref"],"open_access":{"is_oa":true,"oa_status":"gold","oa_url":"https://aclanthology.org/2025.findings-acl.1237.pdf","any_repository_has_fulltext":null},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5100774764","display_name":"Chaojie Wang","orcid":"https://orcid.org/0000-0002-7644-7621"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Chaojie Wang","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":null,"display_name":"Haonan Shi","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Haonan Shi","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5054591923","display_name":"Long Tian","orcid":"https://orcid.org/0000-0002-5069-0973"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Long Tian","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5017743551","display_name":"Bo An","orcid":"https://orcid.org/0000-0002-7064-7438"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Bo An","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5100381753","display_name":"Shuicheng Yan","orcid":"https://orcid.org/0000-0001-8906-3777"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Shuicheng Yan","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]}],"institutions":[],"countries_distinct_count":0,"institutions_distinct_count":5,"corresponding_author_ids":[],"corresponding_institution_ids":[],"apc_list":null,"apc_paid":null,"fwci":0.0,"has_fulltext":true,"cited_by_count":0,"citation_normalized_percentile":{"value":0.08841314,"is_in_top_1_percent":false,"is_in_top_10_percent":false},"cited_by_percentile_year":null,"biblio":{"volume":null,"issue":null,"first_page":"24110","last_page":"24122"},"is_retracted":false,"is_paratext":false,"is_xpac":false,"primary_topic":{"id":"https://openalex.org/T11975","display_name":"Evolutionary Algorithms and Applications","score":0.8009999990463257,"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/T11975","display_name":"Evolutionary Algorithms and Applications","score":0.8009999990463257,"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/T10320","display_name":"Neural Networks and Applications","score":0.7457000017166138,"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/T10429","display_name":"EEG and Brain-Computer Interfaces","score":0.7146000266075134,"subfield":{"id":"https://openalex.org/subfields/2805","display_name":"Cognitive Neuroscience"},"field":{"id":"https://openalex.org/fields/28","display_name":"Neuroscience"},"domain":{"id":"https://openalex.org/domains/1","display_name":"Life Sciences"}}],"keywords":[{"id":"https://openalex.org/keywords/reinforcement-learning","display_name":"Reinforcement learning","score":0.7115679383277893},{"id":"https://openalex.org/keywords/computer-science","display_name":"Computer science","score":0.697853684425354},{"id":"https://openalex.org/keywords/reinforcement","display_name":"Reinforcement","score":0.448725163936615},{"id":"https://openalex.org/keywords/artificial-intelligence","display_name":"Artificial intelligence","score":0.4378300905227661},{"id":"https://openalex.org/keywords/psychology","display_name":"Psychology","score":0.14981040358543396},{"id":"https://openalex.org/keywords/social-psychology","display_name":"Social psychology","score":0.09397554397583008}],"concepts":[{"id":"https://openalex.org/C97541855","wikidata":"https://www.wikidata.org/wiki/Q830687","display_name":"Reinforcement learning","level":2,"score":0.7115679383277893},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.697853684425354},{"id":"https://openalex.org/C67203356","wikidata":"https://www.wikidata.org/wiki/Q1321905","display_name":"Reinforcement","level":2,"score":0.448725163936615},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.4378300905227661},{"id":"https://openalex.org/C15744967","wikidata":"https://www.wikidata.org/wiki/Q9418","display_name":"Psychology","level":0,"score":0.14981040358543396},{"id":"https://openalex.org/C77805123","wikidata":"https://www.wikidata.org/wiki/Q161272","display_name":"Social psychology","level":1,"score":0.09397554397583008}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.18653/v1/2025.findings-acl.1237","is_oa":true,"landing_page_url":"https://doi.org/10.18653/v1/2025.findings-acl.1237","pdf_url":"https://aclanthology.org/2025.findings-acl.1237.pdf","source":null,"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Findings of the Association for Computational Linguistics: ACL 2025","raw_type":"proceedings-article"}],"best_oa_location":{"id":"doi:10.18653/v1/2025.findings-acl.1237","is_oa":true,"landing_page_url":"https://doi.org/10.18653/v1/2025.findings-acl.1237","pdf_url":"https://aclanthology.org/2025.findings-acl.1237.pdf","source":null,"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Findings of the Association for Computational Linguistics: ACL 2025","raw_type":"proceedings-article"},"sustainable_development_goals":[{"score":0.6000000238418579,"display_name":"Reduced inequalities","id":"https://metadata.un.org/sdg/10"}],"awards":[],"funders":[],"has_content":{"grobid_xml":true,"pdf":true},"content_urls":{"pdf":"https://content.openalex.org/works/W4412944612.pdf","grobid_xml":"https://content.openalex.org/works/W4412944612.grobid-xml"},"referenced_works_count":0,"referenced_works":[],"related_works":["https://openalex.org/W4391375266","https://openalex.org/W2899084033","https://openalex.org/W2748952813","https://openalex.org/W4310083477","https://openalex.org/W2328553770","https://openalex.org/W2920061524","https://openalex.org/W1977959518","https://openalex.org/W2038908348","https://openalex.org/W2107890255","https://openalex.org/W2106552856"],"abstract_inverted_index":{"Reinforcement":[0],"Learning":[1],"from":[2,52],"Human":[3],"Feedback":[4],"(RLHF)":[5],"has":[6,30],"become":[7],"an":[8],"essential":[9],"technique":[10],"for":[11],"enhancing":[12,180],"pre-trained":[13],"large":[14],"language":[15],"models":[16,37],"(LLMs)":[17],"to":[18,49,99,131,144,172],"generate":[19,100],"responses":[20,101],"that":[21,55,78,156],"align":[22],"with":[23,91,164],"human":[24],"preferences":[25],"and":[26],"societal":[27],"values.Although":[28],"RLHF":[29,107],"shown":[31],"promise,":[32],"the":[33,41,57,69,111,115,133,149,177,181],"training":[34],"of":[35,43,71,110,167,179,183],"reward":[36,44,87,138,146],"(RMs)":[38],"still":[39],"faces":[40],"challenge":[42],"hacking,":[45],"motivating":[46],"recent":[47],"works":[48],"prevent":[50],"RMs":[51,83],"finding":[53],"shortcuts":[54],"bypass":[56],"intended":[58],"optimization":[59],"objectives":[60],"by":[61,82],"identifying":[62],"simplistic":[63],"patterns":[64],"such":[65],"as":[66],"response":[67],"length.Besides":[68],"issue":[70],"length":[72,169],"bias,":[73,170],"our":[74,157],"work":[75],"firstly":[76],"reveals":[77],"prompt-template":[79,134],"bias":[80,135],"learned":[81],"can":[84,141,160],"also":[85],"cause":[86],"hacking":[88],"when":[89],"dealing":[90],"some":[92],"marginal":[93],"samples,":[94],"resulting":[95],"in":[96,102,114,148,176],"LLMs":[97],"preferring":[98],"a":[103,121,173],"specific":[104],"format":[105,112],"after":[106],"fine-tuning,":[108],"regardless":[109],"requested":[113],"prompt.To":[116],"this":[117],"end,":[118],"we":[119,154],"propose":[120],"low-cost":[122],"but":[123],"effective":[124],"method,":[125],"namely":[126],"Prompt":[127],"Bias":[128],"Calibration":[129],"(PBC),":[130],"estimate":[132],"term":[136],"during":[137],"modeling,":[139],"which":[140],"be":[142,161],"utilized":[143],"calibrate":[145],"scores":[147],"following":[150],"RL":[151],"fine-tuning":[152],"process.Then,":[153],"show":[155],"PBC":[158],"method":[159],"flexibly":[162],"combined":[163],"existing":[165],"algorithms":[166],"removing":[168],"leading":[171],"further":[174],"improvement":[175],"aspect":[178],"quality":[182],"generated":[184],"responses.":[185]},"counts_by_year":[],"updated_date":"2026-06-11T09:08:48.828518","created_date":"2025-10-10T00:00:00"}