{"id":"https://openalex.org/W4414959296","doi":"https://doi.org/10.48550/arxiv.2507.02778","title":"Self-Correction Bench: Uncovering and Addressing the Self-Correction Blind Spot in Large Language Models","display_name":"Self-Correction Bench: Uncovering and Addressing the Self-Correction Blind Spot in Large Language Models","publication_year":2025,"publication_date":"2025-07-03","ids":{"openalex":"https://openalex.org/W4414959296","doi":"https://doi.org/10.48550/arxiv.2507.02778"},"language":"en","primary_location":{"id":"pmh:oai:arXiv.org:2507.02778","doi":null,"is_oa":true,"landing_page_url":"https://arxiv.org/abs/2507.02778","pdf_url":"https://arxiv.org/pdf/2507.02778","source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":"2331-8422","issn":["2331-8422"],"is_oa":true,"is_in_doaj":false,"is_core":false,"listed_in":[],"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":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"text"},"volume":null,"type":"preprint","is_traditional":true,"indexed_in":["arxiv","datacite"],"open_access":{"is_oa":true,"oa_status":"green","oa_url":"https://arxiv.org/pdf/2507.02778","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5110118805","display_name":"Ken Tsui","orcid":null,"observed_orcids":[]},"institutions":[],"countries":[],"is_corresponding":true,"raw_author_name":"Tsui, Ken","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]}],"institutions":[],"countries_distinct_count":0,"institutions_distinct_count":0,"corresponding_author_ids":["https://openalex.org/A5110118805"],"corresponding_institution_ids":[],"apc_list":null,"apc_paid":null,"fwci":null,"has_fulltext":true,"cited_by_count":29,"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":0.6464999914169312,"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":0.6464999914169312,"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":0.22939999401569366,"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/T11689","display_name":"Adversarial Robustness in Machine Learning","score":0.06069999933242798,"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/self-correction","display_name":"self-correction","score":0.9459999799728394},{"id":"https://openalex.org/keywords/large-language-models","display_name":"large language models","score":0.972000002861023},{"id":"https://openalex.org/keywords/error-correction","display_name":"error correction","score":0.453000009059906},{"id":"https://openalex.org/keywords/conversational-roles","display_name":"conversational roles","score":0.5630000233650208},{"id":"https://openalex.org/keywords/representation-space","display_name":"representation space","score":0.734000027179718},{"id":"https://openalex.org/keywords/safety-critical-applications","display_name":"safety-critical applications","score":0.5289999842643738}],"concepts":[{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.6583999991416931},{"id":"https://openalex.org/C43214815","wikidata":"https://www.wikidata.org/wiki/Q7310987","display_name":"Reliability (semiconductor)","level":3,"score":0.5687999725341797},{"id":"https://openalex.org/C153701036","wikidata":"https://www.wikidata.org/wiki/Q659974","display_name":"Trustworthiness","level":2,"score":0.5284000039100647},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.4828000068664551},{"id":"https://openalex.org/C137293760","wikidata":"https://www.wikidata.org/wiki/Q3621696","display_name":"Language model","level":2,"score":0.46889999508857727},{"id":"https://openalex.org/C119857082","wikidata":"https://www.wikidata.org/wiki/Q2539","display_name":"Machine learning","level":1,"score":0.4269999861717224},{"id":"https://openalex.org/C2780009758","wikidata":"https://www.wikidata.org/wiki/Q6804172","display_name":"Measure (data warehouse)","level":2,"score":0.3741999864578247},{"id":"https://openalex.org/C97541855","wikidata":"https://www.wikidata.org/wiki/Q830687","display_name":"Reinforcement learning","level":2,"score":0.35580000281333923},{"id":"https://openalex.org/C64731932","wikidata":"https://www.wikidata.org/wiki/Q371090","display_name":"Blind spot","level":2,"score":0.33970001339912415},{"id":"https://openalex.org/C81917197","wikidata":"https://www.wikidata.org/wiki/Q628760","display_name":"Selection (genetic algorithm)","level":2,"score":0.31529998779296875},{"id":"https://openalex.org/C112930515","wikidata":"https://www.wikidata.org/wiki/Q4389547","display_name":"Risk analysis (engineering)","level":1,"score":0.3140999972820282},{"id":"https://openalex.org/C204321447","wikidata":"https://www.wikidata.org/wiki/Q30642","display_name":"Natural language processing","level":1,"score":0.3125},{"id":"https://openalex.org/C111335779","wikidata":"https://www.wikidata.org/wiki/Q3454686","display_name":"Reduction (mathematics)","level":2,"score":0.31029999256134033},{"id":"https://openalex.org/C18762648","wikidata":"https://www.wikidata.org/wiki/Q42213","display_name":"Work (physics)","level":2,"score":0.3100000023841858},{"id":"https://openalex.org/C26517878","wikidata":"https://www.wikidata.org/wiki/Q228039","display_name":"Key (lock)","level":2,"score":0.30730000138282776},{"id":"https://openalex.org/C180747234","wikidata":"https://www.wikidata.org/wiki/Q23373","display_name":"Cognitive psychology","level":1,"score":0.28769999742507935},{"id":"https://openalex.org/C61797465","wikidata":"https://www.wikidata.org/wiki/Q1188986","display_name":"Term (time)","level":2,"score":0.2711000144481659},{"id":"https://openalex.org/C191147762","wikidata":"https://www.wikidata.org/wiki/Q186289","display_name":"Human reliability","level":3,"score":0.26649999618530273},{"id":"https://openalex.org/C169806903","wikidata":"https://www.wikidata.org/wiki/Q5937752","display_name":"Human error","level":2,"score":0.2614000141620636},{"id":"https://openalex.org/C15744967","wikidata":"https://www.wikidata.org/wiki/Q9418","display_name":"Psychology","level":0,"score":0.2606000006198883},{"id":"https://openalex.org/C103088060","wikidata":"https://www.wikidata.org/wiki/Q1062839","display_name":"Error detection and correction","level":2,"score":0.2529999911785126},{"id":"https://openalex.org/C148220186","wikidata":"https://www.wikidata.org/wiki/Q7111912","display_name":"Outcome (game theory)","level":2,"score":0.2517000138759613},{"id":"https://openalex.org/C51632099","wikidata":"https://www.wikidata.org/wiki/Q3985153","display_name":"Training set","level":2,"score":0.25119999051094055},{"id":"https://openalex.org/C108583219","wikidata":"https://www.wikidata.org/wiki/Q197536","display_name":"Deep learning","level":2,"score":0.2508000135421753}],"mesh":[],"locations_count":2,"locations":[{"id":"pmh:oai:arXiv.org:2507.02778","doi":null,"is_oa":true,"landing_page_url":"https://arxiv.org/abs/2507.02778","pdf_url":"https://arxiv.org/pdf/2507.02778","source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":"2331-8422","issn":["2331-8422"],"is_oa":true,"is_in_doaj":false,"is_core":false,"listed_in":[],"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":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"text"},{"id":"doi:10.48550/arxiv.2507.02778","doi":"https://doi.org/10.48550/arxiv.2507.02778","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2507.02778","pdf_url":null,"source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":"2331-8422","issn":["2331-8422"],"is_oa":true,"is_in_doaj":false,"is_core":false,"listed_in":[],"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"}],"best_oa_location":{"id":"pmh:oai:arXiv.org:2507.02778","doi":null,"is_oa":true,"landing_page_url":"https://arxiv.org/abs/2507.02778","pdf_url":"https://arxiv.org/pdf/2507.02778","source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":"2331-8422","issn":["2331-8422"],"is_oa":true,"is_in_doaj":false,"is_core":false,"listed_in":[],"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":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"text"},"sustainable_development_goals":[],"study_designs":[],"awards":[],"funders":[],"has_content":{"pdf":true,"grobid_xml":true},"content_urls":{"pdf":"https://content.openalex.org/works/W4414959296.pdf","grobid_xml":"https://content.openalex.org/works/W4414959296.grobid-xml"},"referenced_works_count":29,"referenced_works":["https://openalex.org/W2036469702","https://openalex.org/W2979826702","https://openalex.org/W3021338988","https://openalex.org/W3168867926","https://openalex.org/W3210904070","https://openalex.org/W4386081218","https://openalex.org/W4387355345","https://openalex.org/W4387355948","https://openalex.org/W4388092811","https://openalex.org/W4392669753","https://openalex.org/W4399401573","https://openalex.org/W4401660192","https://openalex.org/W4402670356","https://openalex.org/W4402670679","https://openalex.org/W4402671231","https://openalex.org/W4404131920","https://openalex.org/W4404534210","https://openalex.org/W4404780679","https://openalex.org/W4404781852","https://openalex.org/W4405354744","https://openalex.org/W4405655184","https://openalex.org/W4406975744","https://openalex.org/W4412886965","https://openalex.org/W4414875567","https://openalex.org/W4414939355","https://openalex.org/W4415800641","https://openalex.org/W4416036482","https://openalex.org/W4416131312","https://openalex.org/W6910566006"],"related_works":["https://openalex.org/W4399400998","https://openalex.org/W7127318121","https://openalex.org/W7163708183","https://openalex.org/W4412886705","https://openalex.org/W7119018651","https://openalex.org/W4417537343","https://openalex.org/W7161742608","https://openalex.org/W7131822610","https://openalex.org/W4400025607","https://openalex.org/W4392019756"],"abstract_inverted_index":{"Although":[0],"large":[1],"language":[2],"models":[3,86,93],"(LLMs)":[4],"have":[5],"transformed":[6],"AI,":[7],"they":[8],"still":[9],"make":[10],"errors":[11,96],"and":[12,154,196],"follow":[13],"unproductive":[14],"reasoning":[15],"paths.":[16],"Self-correction":[17],"is":[18,41,108,131,137,207],"vital":[19],"for":[20],"safety-critical":[21],"applications,":[22],"but":[23,97,107],"studying":[24],"it":[25,42,44,48],"requires":[26,186],"disentangling":[27],"activation":[28,206],"failure":[29],"from":[30],"knowledge":[31],"deficiency:":[32],"when":[33,133],"a":[34,55,88,117,122,173,199],"model":[35,123],"fails":[36,124],"to":[37,125,144,210],"correct":[38,94],"an":[39,70],"error,":[40,76],"because":[43,47],"cannot,":[45],"or":[46,73],"does":[49],"not?":[50],"We":[51,140],"introduce":[52],"Self-Correction":[53,90],"Bench,":[54],"controlled":[56],"evaluation":[57],"framework":[58],"that":[59,180,204],"isolates":[60],"this":[61,211],"distinction":[62],"by":[63,168,194],"injecting":[64],"the":[65,104,134,142,165,191],"same":[66],"error":[67,136],"as":[68,157,159],"either":[69],"external":[71,95],"(user-attributed)":[72],"internal":[74,101],"(model-attributed)":[75],"keeping":[77],"all":[78],"other":[79],"context":[80],"identical.":[81],"Testing":[82],"14":[83],"open-source":[84],"non-reasoning":[85],"reveals":[87],"64.5%":[89],"Blind":[91],"Spot:":[92],"fail":[98],"on":[99],"identical":[100,135],"ones,":[102],"proving":[103],"capability":[105],"exists":[106],"not":[109,208],"activated.":[110],"On":[111],"models'":[112],"own":[113,129],"naturally":[114],"generated":[115],"errors,":[116],"measurable":[118],"share":[119],"of":[120],"what":[121],"catch":[126],"in":[127,177],"its":[128],"output":[130],"caught":[132],"presented":[138],"externally.":[139],"trace":[141],"cause":[143],"post-training":[145],"data":[146],"composition:":[147],"supervised":[148],"fine-tuning":[149,155],"datasets":[150],"lack":[151],"error-correction":[152],"sequences,":[153],"with":[156],"few":[158],"5,306":[160],"such":[161],"traces":[162],"already":[163],"reduces":[164,190],"blind":[166,192],"spot":[167,193],"76.0%.":[169],"Mechanistically,":[170],"we":[171],"identify":[172],"transferable":[174],"conversational-role":[175],"direction":[176],"representation":[178],"space":[179],"causally":[181],"gates":[182],"self-correction.":[183],"Appending":[184],"\"Wait\"":[185],"no":[187],"training":[188],"yet":[189],"89.3%,":[195],"operates":[197],"through":[198],"nearly":[200],"independent":[201],"pathway,":[202],"indicating":[203],"correction":[205],"reducible":[209],"single":[212],"mechanism.":[213]},"counts_by_year":[{"year":2026,"cited_by_count":28},{"year":2025,"cited_by_count":1}],"updated_date":"2026-11-07T11:25:07.637180","created_date":"2025-10-09T00:00:00"}