{"id":"https://openalex.org/W7163198652","doi":"https://doi.org/10.48550/arxiv.2606.01436","title":"Learning from Saturated Data: Signals Beyond Correctness for LLM Training","display_name":"Learning from Saturated Data: Signals Beyond Correctness for LLM Training","publication_year":2026,"publication_date":"2026-05-31","ids":{"openalex":"https://openalex.org/W7163198652","doi":"https://doi.org/10.48550/arxiv.2606.01436"},"language":null,"primary_location":{"id":"doi:10.48550/arxiv.2606.01436","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2606.01436","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":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":null,"is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"Preprint"},"type":"preprint","indexed_in":["datacite"],"open_access":{"is_oa":true,"oa_status":"green","oa_url":"https://doi.org/10.48550/arxiv.2606.01436","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5137659187","display_name":"Hanno Hiss","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Hiss, Hanno","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5093356429","display_name":"Jasper Dekoninck","orcid":"https://orcid.org/0009-0009-0621-170X"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Dekoninck, Jasper","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5126932701","display_name":"Martin Vechev","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Vechev, Martin","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]}],"institutions":[],"countries_distinct_count":0,"institutions_distinct_count":0,"corresponding_author_ids":[],"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":0.4327999949455261,"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":0.4327999949455261,"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":0.06560000032186508,"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/T11307","display_name":"Domain Adaptation and Few-Shot Learning","score":0.061400000005960464,"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/correctness","display_name":"Correctness","score":0.9023000001907349},{"id":"https://openalex.org/keywords/pairwise-comparison","display_name":"Pairwise comparison","score":0.6643999814987183},{"id":"https://openalex.org/keywords/base","display_name":"Base (topology)","score":0.5598999857902527},{"id":"https://openalex.org/keywords/quality","display_name":"Quality (philosophy)","score":0.5065000057220459},{"id":"https://openalex.org/keywords/binary-number","display_name":"Binary number","score":0.5062000155448914},{"id":"https://openalex.org/keywords/training-set","display_name":"Training set","score":0.4528999924659729},{"id":"https://openalex.org/keywords/training","display_name":"Training (meteorology)","score":0.43700000643730164}],"concepts":[{"id":"https://openalex.org/C55439883","wikidata":"https://www.wikidata.org/wiki/Q360812","display_name":"Correctness","level":2,"score":0.9023000001907349},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.7189000248908997},{"id":"https://openalex.org/C184898388","wikidata":"https://www.wikidata.org/wiki/Q1435712","display_name":"Pairwise comparison","level":2,"score":0.6643999814987183},{"id":"https://openalex.org/C42058472","wikidata":"https://www.wikidata.org/wiki/Q810214","display_name":"Base (topology)","level":2,"score":0.5598999857902527},{"id":"https://openalex.org/C119857082","wikidata":"https://www.wikidata.org/wiki/Q2539","display_name":"Machine learning","level":1,"score":0.5198000073432922},{"id":"https://openalex.org/C2779530757","wikidata":"https://www.wikidata.org/wiki/Q1207505","display_name":"Quality (philosophy)","level":2,"score":0.5065000057220459},{"id":"https://openalex.org/C48372109","wikidata":"https://www.wikidata.org/wiki/Q3913","display_name":"Binary number","level":2,"score":0.5062000155448914},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.5049999952316284},{"id":"https://openalex.org/C51632099","wikidata":"https://www.wikidata.org/wiki/Q3985153","display_name":"Training set","level":2,"score":0.4528999924659729},{"id":"https://openalex.org/C2777211547","wikidata":"https://www.wikidata.org/wiki/Q17141490","display_name":"Training (meteorology)","level":2,"score":0.43700000643730164},{"id":"https://openalex.org/C165838908","wikidata":"https://www.wikidata.org/wiki/Q736777","display_name":"Calibration","level":2,"score":0.40939998626708984},{"id":"https://openalex.org/C11413529","wikidata":"https://www.wikidata.org/wiki/Q8366","display_name":"Algorithm","level":1,"score":0.3993000090122223},{"id":"https://openalex.org/C2780586882","wikidata":"https://www.wikidata.org/wiki/Q7520643","display_name":"Simple (philosophy)","level":2,"score":0.37310001254081726},{"id":"https://openalex.org/C2779843651","wikidata":"https://www.wikidata.org/wiki/Q7390335","display_name":"SIGNAL (programming language)","level":2,"score":0.3564999997615814},{"id":"https://openalex.org/C124101348","wikidata":"https://www.wikidata.org/wiki/Q172491","display_name":"Data mining","level":1,"score":0.33009999990463257},{"id":"https://openalex.org/C2780148112","wikidata":"https://www.wikidata.org/wiki/Q1432581","display_name":"Proxy (statistics)","level":2,"score":0.3280999958515167},{"id":"https://openalex.org/C104267543","wikidata":"https://www.wikidata.org/wiki/Q208163","display_name":"Signal processing","level":3,"score":0.31690001487731934},{"id":"https://openalex.org/C120936955","wikidata":"https://www.wikidata.org/wiki/Q2155640","display_name":"Empirical research","level":2,"score":0.2775999903678894},{"id":"https://openalex.org/C66905080","wikidata":"https://www.wikidata.org/wiki/Q17005494","display_name":"Binary classification","level":3,"score":0.2687000036239624},{"id":"https://openalex.org/C35639132","wikidata":"https://www.wikidata.org/wiki/Q7452468","display_name":"Sequence labeling","level":3,"score":0.26669999957084656},{"id":"https://openalex.org/C28490314","wikidata":"https://www.wikidata.org/wiki/Q189436","display_name":"Speech recognition","level":1,"score":0.26660001277923584}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.48550/arxiv.2606.01436","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2606.01436","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":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":null,"is_accepted":false,"is_published":null,"raw_source_name":null,"raw_type":"Preprint"}],"best_oa_location":{"id":"doi:10.48550/arxiv.2606.01436","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2606.01436","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":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":null,"is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"Preprint"},"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":{"The":[0],"growing":[1],"capabilities":[2],"of":[3,13,53,70],"large":[4],"language":[5],"models":[6],"(LLMs)":[7],"have":[8],"led":[9],"to":[10,20,39,116,181],"the":[11,64,67,119,136,156],"saturation":[12],"many":[14],"benchmarks":[15],"and":[16,74,97,132,150,188],"training":[17,95,103,165],"datasets":[18],"used":[19,38,82],"improve":[21,40,112],"them.":[22],"Motivated":[23],"by":[24,114],"this,":[25],"we":[26,46],"investigate":[27],"whether":[28],"questions":[29,172],"solved":[30],"with":[31,50,145],"perfect":[32],"empirical":[33],"accuracy":[34],"can":[35,151,166],"nevertheless":[36],"be":[37],"downstream":[41],"performance.":[42],"To":[43],"do":[44],"so,":[45],"replace":[47],"binary":[48],"correctness":[49],"two":[51],"sources":[52],"more":[54,130,182],"fine-grained":[55],"quality":[56,69,137],"signals:":[57],"(1)":[58],"pairwise":[59],"LLM":[60],"self-judgments,":[61],"in":[62],"which":[63],"model":[65],"evaluates":[66],"relative":[68],"its":[71],"own":[72],"solutions,":[73],"(2)":[75],"token-level":[76,79],"entropy,":[77],"where":[78],"uncertainty":[80],"is":[81],"as":[83],"a":[84,106,146],"proxy":[85],"for":[86,173],"solution":[87],"quality.":[88],"We":[89],"incorporate":[90],"these":[91],"signals":[92,111,180],"into":[93],"several":[94],"algorithms":[96],"evaluate":[98],"them":[99],"on":[100,105,135],"Qwen3-1.7B-Base.":[101],"When":[102],"exclusively":[104],"simple":[107],"arithmetic":[108],"task,":[109],"quality-based":[110,164],"performance":[113,154],"up":[115],"$18.6\\%$":[117],"over":[118],"base":[120,157,174],"model,":[121],"substantially":[122],"outperforming":[123],"SFT.":[124],"On":[125],"GSM8K,":[126],"however,":[127],"gains":[128],"are":[129],"modest":[131],"depend":[133],"strongly":[134],"signal.":[138],"For":[139],"instance,":[140],"self-judgments":[141],"show":[142],"poor":[143],"agreement":[144],"stronger":[147],"external":[148],"judge":[149],"even":[152],"degrade":[153],"below":[155],"model.":[158],"Overall,":[159],"our":[160],"results":[161],"suggest":[162],"that":[163,177],"extract":[167],"useful":[168],"signal":[169],"from":[170],"saturated":[171],"models,":[175],"but":[176],"applying":[178],"such":[179],"complex":[183],"tasks":[184],"requires":[185],"careful":[186],"calibration":[187],"further":[189],"study.":[190]},"counts_by_year":[],"updated_date":"2026-07-01T06:00:48.157686","created_date":"2026-06-03T00:00:00"}