{"id":"https://openalex.org/W7133557317","doi":"https://doi.org/10.48550/arxiv.2603.02225","title":"Scaling Reward Modeling without Human Supervision","display_name":"Scaling Reward Modeling without Human Supervision","publication_year":2026,"publication_date":"2026-02-11","ids":{"openalex":"https://openalex.org/W7133557317","doi":"https://doi.org/10.48550/arxiv.2603.02225"},"language":null,"primary_location":{"id":"doi:10.48550/arxiv.2603.02225","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2603.02225","pdf_url":null,"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":null,"is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"article"},"type":"preprint","indexed_in":["datacite"],"open_access":{"is_oa":true,"oa_status":"green","oa_url":"https://doi.org/10.48550/arxiv.2603.02225","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5057660228","display_name":"Jingxuan Fan","orcid":"https://orcid.org/0000-0001-7860-4558"},"institutions":[],"countries":[],"is_corresponding":true,"raw_author_name":"Fan, Jingxuan","raw_affiliation_strings":[],"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5021113995","display_name":"Y. Li","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Li, Yueying","raw_affiliation_strings":[],"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5128123368","display_name":"Zhenting Qi","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Qi, Zhenting","raw_affiliation_strings":[],"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5128110748","display_name":"Dinghuai Zhang","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Zhang, Dinghuai","raw_affiliation_strings":[],"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5022968100","display_name":"Kiant\u00e9 Brantley","orcid":"https://orcid.org/0000-0002-8395-594X"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Brantley, Kiant\u00e9","raw_affiliation_strings":[],"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5022501668","display_name":"Sham Kakade","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Kakade, Sham M.","raw_affiliation_strings":[],"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5128061493","display_name":"Hanlin Zhang","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Zhang, Hanlin","raw_affiliation_strings":[],"affiliations":[]}],"institutions":[],"countries_distinct_count":0,"institutions_distinct_count":7,"corresponding_author_ids":["https://openalex.org/A5057660228"],"corresponding_institution_ids":[],"apc_list":null,"apc_paid":null,"fwci":null,"has_fulltext":false,"cited_by_count":0,"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/T10028","display_name":"Topic Modeling","score":0.2912999987602234,"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":0.2912999987602234,"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":0.13410000503063202,"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/T12535","display_name":"Machine Learning and Data Classification","score":0.1281999945640564,"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/process","display_name":"Process (computing)","score":0.521399974822998},{"id":"https://openalex.org/keywords/selection","display_name":"Selection (genetic algorithm)","score":0.43389999866485596},{"id":"https://openalex.org/keywords/operationalization","display_name":"Operationalization","score":0.41269999742507935},{"id":"https://openalex.org/keywords/scaling","display_name":"Scaling","score":0.39410001039505005},{"id":"https://openalex.org/keywords/initialization","display_name":"Initialization","score":0.3790000081062317},{"id":"https://openalex.org/keywords/model-selection","display_name":"Model selection","score":0.3528999984264374},{"id":"https://openalex.org/keywords/preference","display_name":"Preference","score":0.3246000111103058}],"concepts":[{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.6959999799728394},{"id":"https://openalex.org/C98045186","wikidata":"https://www.wikidata.org/wiki/Q205663","display_name":"Process (computing)","level":2,"score":0.521399974822998},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.4489000141620636},{"id":"https://openalex.org/C119857082","wikidata":"https://www.wikidata.org/wiki/Q2539","display_name":"Machine learning","level":1,"score":0.4453999996185303},{"id":"https://openalex.org/C81917197","wikidata":"https://www.wikidata.org/wiki/Q628760","display_name":"Selection (genetic algorithm)","level":2,"score":0.43389999866485596},{"id":"https://openalex.org/C9354725","wikidata":"https://www.wikidata.org/wiki/Q286017","display_name":"Operationalization","level":2,"score":0.41269999742507935},{"id":"https://openalex.org/C99844830","wikidata":"https://www.wikidata.org/wiki/Q102441924","display_name":"Scaling","level":2,"score":0.39410001039505005},{"id":"https://openalex.org/C114466953","wikidata":"https://www.wikidata.org/wiki/Q6034165","display_name":"Initialization","level":2,"score":0.3790000081062317},{"id":"https://openalex.org/C93959086","wikidata":"https://www.wikidata.org/wiki/Q6888345","display_name":"Model selection","level":2,"score":0.3528999984264374},{"id":"https://openalex.org/C2781249084","wikidata":"https://www.wikidata.org/wiki/Q908656","display_name":"Preference","level":2,"score":0.3246000111103058},{"id":"https://openalex.org/C160234255","wikidata":"https://www.wikidata.org/wiki/Q812535","display_name":"Bayesian inference","level":3,"score":0.28349998593330383},{"id":"https://openalex.org/C2777211547","wikidata":"https://www.wikidata.org/wiki/Q17141490","display_name":"Training (meteorology)","level":2,"score":0.2727000117301941},{"id":"https://openalex.org/C101097943","wikidata":"https://www.wikidata.org/wiki/Q5176983","display_name":"Counterintuitive","level":2,"score":0.26570001244544983},{"id":"https://openalex.org/C62230096","wikidata":"https://www.wikidata.org/wiki/Q275969","display_name":"Crowdsourcing","level":2,"score":0.2655999958515167},{"id":"https://openalex.org/C34585555","wikidata":"https://www.wikidata.org/wiki/Q1368723","display_name":"Learning curve","level":2,"score":0.2605000138282776},{"id":"https://openalex.org/C58973888","wikidata":"https://www.wikidata.org/wiki/Q1041418","display_name":"Semi-supervised learning","level":2,"score":0.2549999952316284},{"id":"https://openalex.org/C107673813","wikidata":"https://www.wikidata.org/wiki/Q812534","display_name":"Bayesian probability","level":2,"score":0.25360000133514404}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.48550/arxiv.2603.02225","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2603.02225","pdf_url":null,"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":null,"is_accepted":false,"is_published":null,"raw_source_name":null,"raw_type":"article"}],"best_oa_location":{"id":"doi:10.48550/arxiv.2603.02225","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2603.02225","pdf_url":null,"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":null,"is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"article"},"sustainable_development_goals":[],"awards":[],"funders":[],"has_content":{"grobid_xml":false,"pdf":false},"content_urls":null,"referenced_works_count":0,"referenced_works":[],"related_works":[],"abstract_inverted_index":{"Learning":[0],"from":[1,57],"feedback":[2],"is":[3,19,63],"an":[4],"instrumental":[5],"process":[6],"for":[7],"advancing":[8],"the":[9,168],"capabilities":[10],"and":[11,24,54,87,89,101,128,134,142,153,170,178],"safety":[12,133],"of":[13,77,120,162,172],"frontier":[14],"models,":[15,104],"yet":[16],"its":[17,45],"effectiveness":[18],"often":[20],"constrained":[21],"by":[22,111],"cost":[23],"scalability.":[25],"We":[26,39],"present":[27],"a":[28],"pilot":[29],"study":[30],"that":[31],"explores":[32],"scaling":[33,42],"reward":[34,146,159,174],"models":[35,147,175],"through":[36],"unsupervised":[37],"approaches.":[38],"operationalize":[40],"reward-based":[41],"(RBS),":[43],"in":[44,65],"simplest":[46],"form,":[47],"as":[48],"preference":[49],"learning":[50],"over":[51],"document":[52],"prefixes":[53],"suffixes":[55],"drawn":[56],"large-scale":[58],"web":[59,79],"corpora.":[60],"Its":[61],"advantage":[62],"demonstrated":[64],"various":[66],"aspects:":[67],"despite":[68],"using":[69],"no":[70],"human":[71,181],"annotations,":[72],"training":[73,173],"on":[74,84,116,124,131],"11M":[75],"tokens":[76],"math-focused":[78],"data":[80],"yields":[81],"steady":[82],"gains":[83,119],"RewardBench":[85,108],"v1":[86],"v2,":[88],"these":[90,145],"improvements":[91,130],"consistently":[92],"transfer":[93],"across":[94],"diverse":[95],"initialization":[96],"backbones":[97],"spanning":[98],"model":[99,160],"families":[100],"scales.":[102],"Across":[103],"our":[105],"method":[106],"improves":[107],"v2":[109],"accuracy":[110],"up":[112,121],"to":[113,122,139],"+7.7":[114],"points":[115],"average,":[117],"with":[118],"+16.1":[123],"in-domain":[125],"math":[126,151],"subsets":[127],"consistent":[129],"out-of-domain":[132],"general":[135],"subsets.":[136],"When":[137],"applied":[138],"best-of-N":[140],"selection":[141],"policy":[143],"optimization,":[144],"substantially":[148],"improve":[149],"downstream":[150],"performance":[152],"match":[154],"or":[155],"exceed":[156],"strong":[157],"supervised":[158],"baselines":[161],"similar":[163],"size.":[164],"Overall,":[165],"we":[166],"demonstrate":[167],"feasibility":[169],"promise":[171],"without":[176],"costly":[177],"potentially":[179],"unreliable":[180],"annotations.":[182]},"counts_by_year":[],"updated_date":"2026-03-05T07:36:02.291473","created_date":"2026-03-05T00:00:00"}