{"id":"https://openalex.org/W7161988363","doi":"https://doi.org/10.48550/arxiv.2605.20285","title":"Introspective X Training: Feedback Conditioning Improves Scaling Across all LLM Training Stages","display_name":"Introspective X Training: Feedback Conditioning Improves Scaling Across all LLM Training Stages","publication_year":2026,"publication_date":"2026-05-19","ids":{"openalex":"https://openalex.org/W7161988363","doi":"https://doi.org/10.48550/arxiv.2605.20285"},"language":null,"primary_location":{"id":"doi:10.48550/arxiv.2605.20285","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2605.20285","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,"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.2605.20285","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5067671433","display_name":"Brandon Cui","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Cui, Brandon","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5136635290","display_name":"Ximing Lu","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Lu, Ximing","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5136637967","display_name":"Jaehun Jung","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Jung, Jaehun","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5112924096","display_name":"Syeda Nahida Akter","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Akter, Syeda Nahida","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5136719099","display_name":"Hyunwoo Kim","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Kim, Hyunwoo","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5136682660","display_name":"Yuxiao Qu","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Qu, Yuxiao","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5030331254","display_name":"David Acuna","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Acuna, David","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5136669907","display_name":"Shrimai Prabhumoye","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Prabhumoye, Shrimai","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5136648552","display_name":"Yejin Choi","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Choi, Yejin","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5029637639","display_name":"Prithviraj Ammanabrolu","orcid":"https://orcid.org/0000-0003-1950-0454"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Ammanabrolu, Prithviraj","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":0.2379000037908554,"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":0.2379000037908554,"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/T11714","display_name":"Multimodal Machine Learning Applications","score":0.08910000324249268,"subfield":{"id":"https://openalex.org/subfields/1707","display_name":"Computer Vision and Pattern Recognition"},"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":0.07029999792575836,"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/introspection","display_name":"Introspection","score":0.784500002861023},{"id":"https://openalex.org/keywords/intuition","display_name":"Intuition","score":0.6279000043869019},{"id":"https://openalex.org/keywords/training","display_name":"Training (meteorology)","score":0.5651000142097473},{"id":"https://openalex.org/keywords/scaling","display_name":"Scaling","score":0.5504000186920166},{"id":"https://openalex.org/keywords/scratch","display_name":"Scratch","score":0.5216000080108643},{"id":"https://openalex.org/keywords/quality","display_name":"Quality (philosophy)","score":0.4368000030517578}],"concepts":[{"id":"https://openalex.org/C129671850","wikidata":"https://www.wikidata.org/wiki/Q210501","display_name":"Introspection","level":2,"score":0.784500002861023},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.6751000285148621},{"id":"https://openalex.org/C132010649","wikidata":"https://www.wikidata.org/wiki/Q189222","display_name":"Intuition","level":2,"score":0.6279000043869019},{"id":"https://openalex.org/C2777211547","wikidata":"https://www.wikidata.org/wiki/Q17141490","display_name":"Training (meteorology)","level":2,"score":0.5651000142097473},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.5613999962806702},{"id":"https://openalex.org/C99844830","wikidata":"https://www.wikidata.org/wiki/Q102441924","display_name":"Scaling","level":2,"score":0.5504000186920166},{"id":"https://openalex.org/C2781235140","wikidata":"https://www.wikidata.org/wiki/Q275131","display_name":"Scratch","level":2,"score":0.5216000080108643},{"id":"https://openalex.org/C2779530757","wikidata":"https://www.wikidata.org/wiki/Q1207505","display_name":"Quality (philosophy)","level":2,"score":0.4368000030517578},{"id":"https://openalex.org/C97541855","wikidata":"https://www.wikidata.org/wiki/Q830687","display_name":"Reinforcement learning","level":2,"score":0.37369999289512634},{"id":"https://openalex.org/C2778755073","wikidata":"https://www.wikidata.org/wiki/Q10858537","display_name":"Scale (ratio)","level":2,"score":0.35429999232292175},{"id":"https://openalex.org/C180747234","wikidata":"https://www.wikidata.org/wiki/Q23373","display_name":"Cognitive psychology","level":1,"score":0.3393999934196472},{"id":"https://openalex.org/C119857082","wikidata":"https://www.wikidata.org/wiki/Q2539","display_name":"Machine learning","level":1,"score":0.31779998540878296},{"id":"https://openalex.org/C107457646","wikidata":"https://www.wikidata.org/wiki/Q207434","display_name":"Human\u2013computer interaction","level":1,"score":0.2955000102519989},{"id":"https://openalex.org/C51632099","wikidata":"https://www.wikidata.org/wiki/Q3985153","display_name":"Training set","level":2,"score":0.29510000348091125},{"id":"https://openalex.org/C15744967","wikidata":"https://www.wikidata.org/wiki/Q9418","display_name":"Psychology","level":0,"score":0.2937999963760376},{"id":"https://openalex.org/C175154964","wikidata":"https://www.wikidata.org/wiki/Q380077","display_name":"Task analysis","level":3,"score":0.27889999747276306},{"id":"https://openalex.org/C3020716817","wikidata":"https://www.wikidata.org/wiki/Q4132092","display_name":"Visual feedback","level":2,"score":0.2639999985694885},{"id":"https://openalex.org/C157742956","wikidata":"https://www.wikidata.org/wiki/Q3237776","display_name":"Frequency scaling","level":3,"score":0.2522999942302704}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.48550/arxiv.2605.20285","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2605.20285","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,"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.2605.20285","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2605.20285","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,"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":[{"id":"https://metadata.un.org/sdg/4","display_name":"Quality Education","score":0.786268413066864}],"awards":[],"funders":[],"has_content":{"grobid_xml":false,"pdf":false},"content_urls":null,"referenced_works_count":0,"referenced_works":[],"related_works":[],"abstract_inverted_index":{"We":[0],"tackle":[1],"the":[2,11,25,28,34,90,94,102,105,135],"question":[3],"of":[4,15,30,33,68,93],"how":[5],"to":[6,41,65,76,138,153],"scale":[7],"more":[8,155],"efficiently":[9],"across":[10],"many,":[12],"ever-growing":[13],"stages":[14,32,44,92],"current":[16],"LLM":[17],"training":[18,88,121],"pipelines.":[19],"Our":[20],"guiding":[21],"intuition":[22],"stems":[23],"from":[24,89,132],"fact":[26],"that":[27,110,144],"dynamics":[29],"later":[31],"pipeline,":[35],"e.g.":[36],"post-training,":[37],"can":[38],"be":[39],"used":[40],"inform":[42],"earlier":[43,119],"such":[45,170],"as":[46,171],"pre-training.":[47],"To":[48],"this":[49],"end,":[50],"we":[51],"propose":[52],"Introspective":[53],"Training":[54],"(or":[55],"IXT),":[56],"inspired":[57],"by":[58,100],"offline":[59],"reward-conditioned":[60],"reinforcement":[61],"learning":[62],"and":[63,159,173],"applicable":[64],"any":[66],"stage":[67],"training.":[69],"IXT":[70],"uses":[71],"a":[72],"thinking":[73],"reward":[74],"model":[75],"annotate":[77],"data":[78,103],"with":[79,104],"natural":[80],"language":[81],"critique":[82],"based":[83],"feedback,":[84],"enabling":[85],"quality":[86],"aware":[87],"earliest":[91],"pipeline.":[95],"Models":[96],"are":[97,114],"then":[98],"trained":[99,131,166],"prefix-conditioning":[101],"generated":[106],"feedback":[107],"--":[108],"ensuring":[109],"not":[111],"all":[112,134],"tokens":[113,141],"treated":[115],"equally":[116],"starting":[117],"much":[118],"in":[120,151,168],"than":[122],"usual.":[123],"Comprehensive":[124],"experiments":[125],"on":[126],"7.5-12B":[127],"transformer-based":[128],"dense":[129],"LLMs":[130],"scratch":[133],"way":[136],"up":[137,152],"18":[139],"Trillion":[140],"seen":[142],"show":[143],"our":[145],"method:":[146],"bends":[147],"scaling":[148],"curves":[149],"resulting":[150],"2.8x":[154],"compute":[156],"efficiency":[157],"generally;":[158],"reaches":[160],"performance":[161],"levels":[162],"unachievable":[163],"for":[164],"models":[165],"otherwise":[167],"domains":[169],"math":[172],"code.":[174]},"counts_by_year":[],"updated_date":"2026-07-28T07:46:37.118299","created_date":"2026-05-22T00:00:00"}