{"id":"https://openalex.org/W4411550453","doi":"https://doi.org/10.1145/3715275.3732068","title":"Reward Model Interpretability via Optimal and Pessimal Tokens","display_name":"Reward Model Interpretability via Optimal and Pessimal Tokens","publication_year":2025,"publication_date":"2025-06-23","ids":{"openalex":"https://openalex.org/W4411550453","doi":"https://doi.org/10.1145/3715275.3732068"},"language":"en","primary_location":{"id":"doi:10.1145/3715275.3732068","is_oa":true,"landing_page_url":"https://doi.org/10.1145/3715275.3732068","pdf_url":null,"source":null,"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 2025 ACM Conference on Fairness, Accountability, and Transparency","raw_type":"proceedings-article"},"type":"preprint","indexed_in":["arxiv","crossref"],"open_access":{"is_oa":true,"oa_status":"gold","oa_url":"https://doi.org/10.1145/3715275.3732068","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5011455884","display_name":"Brian Christian","orcid":"https://orcid.org/0000-0001-5277-8939"},"institutions":[{"id":"https://openalex.org/I40120149","display_name":"University of Oxford","ror":"https://ror.org/052gg0110","country_code":"GB","type":"education","lineage":["https://openalex.org/I40120149"]}],"countries":["GB"],"is_corresponding":false,"raw_author_name":"Brian Christian","raw_affiliation_strings":["University of Oxford, Oxford, United Kingdom"],"raw_orcid":"https://orcid.org/0000-0001-5277-8939","affiliations":[{"raw_affiliation_string":"University of Oxford, Oxford, United Kingdom","institution_ids":["https://openalex.org/I40120149"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5051050719","display_name":"Hannah Rose Kirk","orcid":"https://orcid.org/0000-0002-7419-5993"},"institutions":[{"id":"https://openalex.org/I40120149","display_name":"University of Oxford","ror":"https://ror.org/052gg0110","country_code":"GB","type":"education","lineage":["https://openalex.org/I40120149"]}],"countries":["GB"],"is_corresponding":false,"raw_author_name":"Hannah Rose Kirk","raw_affiliation_strings":["University of Oxford, Oxford, United Kingdom"],"raw_orcid":"https://orcid.org/0000-0002-7419-5993","affiliations":[{"raw_affiliation_string":"University of Oxford, Oxford, United Kingdom","institution_ids":["https://openalex.org/I40120149"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5047952620","display_name":"Jessica A. F. Thompson","orcid":"https://orcid.org/0000-0003-0468-5097"},"institutions":[{"id":"https://openalex.org/I40120149","display_name":"University of Oxford","ror":"https://ror.org/052gg0110","country_code":"GB","type":"education","lineage":["https://openalex.org/I40120149"]}],"countries":["GB"],"is_corresponding":false,"raw_author_name":"Jessica A.F. Thompson","raw_affiliation_strings":["University of Oxford, Oxford, United Kingdom"],"raw_orcid":"https://orcid.org/0000-0003-0468-5097","affiliations":[{"raw_affiliation_string":"University of Oxford, Oxford, United Kingdom","institution_ids":["https://openalex.org/I40120149"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5031878516","display_name":"Christopher Summerfield","orcid":"https://orcid.org/0000-0002-2941-2653"},"institutions":[{"id":"https://openalex.org/I40120149","display_name":"University of Oxford","ror":"https://ror.org/052gg0110","country_code":"GB","type":"education","lineage":["https://openalex.org/I40120149"]}],"countries":["GB"],"is_corresponding":false,"raw_author_name":"Christopher Summerfield","raw_affiliation_strings":["University of Oxford, Oxford, United Kingdom"],"raw_orcid":"https://orcid.org/0000-0002-2941-2653","affiliations":[{"raw_affiliation_string":"University of Oxford, Oxford, United Kingdom","institution_ids":["https://openalex.org/I40120149"]}]},{"author_position":"last","author":{"id":"https://openalex.org/A5045292532","display_name":"Tsvetomira Dumbalska","orcid":"https://orcid.org/0000-0002-5761-8536"},"institutions":[{"id":"https://openalex.org/I40120149","display_name":"University of Oxford","ror":"https://ror.org/052gg0110","country_code":"GB","type":"education","lineage":["https://openalex.org/I40120149"]}],"countries":["GB"],"is_corresponding":false,"raw_author_name":"Tsvetomira Dumbalska","raw_affiliation_strings":["University of Oxford, Oxford, United Kingdom"],"raw_orcid":"https://orcid.org/0000-0002-5761-8536","affiliations":[{"raw_affiliation_string":"University of Oxford, Oxford, United Kingdom","institution_ids":["https://openalex.org/I40120149"]}]}],"institutions":[],"countries_distinct_count":1,"institutions_distinct_count":1,"corresponding_author_ids":[],"corresponding_institution_ids":["https://openalex.org/I40120149"],"apc_list":null,"apc_paid":null,"fwci":5.7896,"has_fulltext":false,"cited_by_count":4,"citation_normalized_percentile":{"value":0.95726755,"is_in_top_1_percent":false,"is_in_top_10_percent":true},"cited_by_percentile_year":{"min":95,"max":97},"biblio":{"volume":null,"issue":null,"first_page":"1048","last_page":"1059"},"is_retracted":false,"is_paratext":false,"is_xpac":false,"primary_topic":{"id":"https://openalex.org/T11689","display_name":"Adversarial Robustness in Machine Learning","score":0.9976999759674072,"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/T11689","display_name":"Adversarial Robustness in Machine Learning","score":0.9976999759674072,"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":0.9832000136375427,"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":0.9822999835014343,"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/interpretability","display_name":"Interpretability","score":0.9132499694824219},{"id":"https://openalex.org/keywords/computer-science","display_name":"Computer science","score":0.7447117567062378},{"id":"https://openalex.org/keywords/artificial-intelligence","display_name":"Artificial intelligence","score":0.40560418367385864},{"id":"https://openalex.org/keywords/natural-language-processing","display_name":"Natural language processing","score":0.3879905939102173}],"concepts":[{"id":"https://openalex.org/C2781067378","wikidata":"https://www.wikidata.org/wiki/Q17027399","display_name":"Interpretability","level":2,"score":0.9132499694824219},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.7447117567062378},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.40560418367385864},{"id":"https://openalex.org/C204321447","wikidata":"https://www.wikidata.org/wiki/Q30642","display_name":"Natural language processing","level":1,"score":0.3879905939102173}],"mesh":[],"locations_count":3,"locations":[{"id":"doi:10.1145/3715275.3732068","is_oa":true,"landing_page_url":"https://doi.org/10.1145/3715275.3732068","pdf_url":null,"source":null,"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 2025 ACM Conference on Fairness, Accountability, and Transparency","raw_type":"proceedings-article"},{"id":"pmh:oai:arXiv.org:2506.07326","is_oa":true,"landing_page_url":"http://arxiv.org/abs/2506.07326","pdf_url":"https://arxiv.org/pdf/2506.07326","source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":null,"issn":null,"is_oa":true,"is_in_doaj":false,"is_core":false,"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":"pmh:doi:10.48550/arxiv.2506.07326","is_oa":true,"landing_page_url":null,"pdf_url":null,"source":{"id":"https://openalex.org/S4406922384","display_name":"Open MIND","issn_l":null,"issn":null,"is_oa":false,"is_in_doaj":false,"is_core":false,"host_organization":null,"host_organization_name":null,"host_organization_lineage":[],"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":"Article"}],"best_oa_location":{"id":"doi:10.1145/3715275.3732068","is_oa":true,"landing_page_url":"https://doi.org/10.1145/3715275.3732068","pdf_url":null,"source":null,"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 2025 ACM Conference on Fairness, Accountability, and Transparency","raw_type":"proceedings-article"},"sustainable_development_goals":[],"awards":[],"funders":[],"has_content":{"pdf":false,"grobid_xml":false},"content_urls":null,"referenced_works_count":21,"referenced_works":["https://openalex.org/W1849277567","https://openalex.org/W1932198206","https://openalex.org/W2013784666","https://openalex.org/W2096648510","https://openalex.org/W2116064496","https://openalex.org/W2157310699","https://openalex.org/W2612690371","https://openalex.org/W2791170418","https://openalex.org/W2949678053","https://openalex.org/W2962990575","https://openalex.org/W2990138404","https://openalex.org/W3165553112","https://openalex.org/W4229494842","https://openalex.org/W4241560494","https://openalex.org/W4242336474","https://openalex.org/W4385573644","https://openalex.org/W4386242333","https://openalex.org/W4389519493","https://openalex.org/W4389523709","https://openalex.org/W4404088618","https://openalex.org/W4404792842"],"related_works":["https://openalex.org/W4391375266","https://openalex.org/W2899084033","https://openalex.org/W2748952813","https://openalex.org/W2905433371","https://openalex.org/W2888392564","https://openalex.org/W4310278675","https://openalex.org/W4388422664","https://openalex.org/W4390569940","https://openalex.org/W4361193272","https://openalex.org/W3204019825"],"abstract_inverted_index":{"Reward":[0],"modeling":[1],"has":[2,18],"emerged":[3],"as":[4,24,157,159,162,186],"a":[5,25,56],"crucial":[6],"component":[7],"in":[8,105],"aligning":[9],"large":[10,200],"language":[11,201],"models":[12,23,34,79,97,107,140,174,202],"with":[13],"human":[14,40,121,168],"values.":[15,169],"Significant":[16],"attention":[17],"focused":[19],"on":[20,99],"using":[21],"reward":[22,33,60,78,139,155],"means":[26],"for":[27],"fine-tuning":[28],"generative":[29],"models.":[30],"However,":[31],"the":[32,152,198],"themselves":[35],"--":[36,50,192],"which":[37,183],"directly":[38],"encode":[39,108,176],"value":[41],"judgments":[42],"by":[43],"turning":[44],"prompt-response":[45],"pairs":[46],"into":[47],"scalar":[48],"rewards":[49],"remain":[51],"relatively":[52],"understudied.":[53],"We":[54,131,170],"present":[55],"novel":[57],"approach":[58],"to":[59,85,116,205],"model":[61],"interpretability":[62],"through":[63,197],"exhaustive":[64],"analysis":[65],"of":[66,127,141,154,164,189],"their":[67,70,160],"responses":[68],"across":[69,135],"entire":[71],"vocabulary":[72],"space.":[73],"By":[74],"examining":[75],"how":[76,106],"different":[77],"score":[80],"every":[81],"possible":[82],"single-token":[83],"response":[84],"value-laden":[86],"prompts,":[87],"we":[88],"uncover":[89],"several":[90],"striking":[91],"findings:":[92],"(i)":[93],"substantial":[94],"heterogeneity":[95],"between":[96],"trained":[98],"similar":[100],"objectives,":[101],"(ii)":[102],"systematic":[103],"asymmetries":[104],"high-":[109],"vs":[110],"low-scoring":[111],"tokens,":[112],"(iii)":[113],"significant":[114],"sensitivity":[115],"prompt":[117],"framing":[118],"that":[119,172,194],"mirrors":[120],"cognitive":[122],"biases,":[123],"and":[124,145,166],"(iv)":[125],"overvaluation":[126],"more":[128],"frequent":[129],"tokens.":[130],"demonstrate":[132],"these":[133,173],"effects":[134],"ten":[136],"recent":[137],"open-source":[138],"varying":[142],"parameter":[143],"counts":[144],"architectures.":[146],"Our":[147],"results":[148],"challenge":[149],"assumptions":[150],"about":[151],"interchangeability":[153],"models,":[156],"well":[158],"suitability":[161],"proxies":[163],"complex":[165],"context-dependent":[167],"find":[171],"can":[175],"concerning":[177],"biases":[178],"toward":[179],"certain":[180],"identity":[181],"groups,":[182],"may":[184],"emerge":[185],"unintended":[187],"consequences":[188],"harmlessness":[190],"training":[191],"distortions":[193],"risk":[195],"propagating":[196],"downstream":[199],"now":[203],"deployed":[204],"millions.":[206]},"counts_by_year":[{"year":2026,"cited_by_count":1},{"year":2025,"cited_by_count":3}],"updated_date":"2026-07-03T08:13:44.112507","created_date":"2025-10-10T00:00:00"}