{"id":"https://openalex.org/W7160938662","doi":"https://doi.org/10.48550/arxiv.2605.09995","title":"Annotations Mitigate Post-Training Mode Collapse","display_name":"Annotations Mitigate Post-Training Mode Collapse","publication_year":2026,"publication_date":"2026-05-11","ids":{"openalex":"https://openalex.org/W7160938662","doi":"https://doi.org/10.48550/arxiv.2605.09995"},"language":null,"primary_location":{"id":"doi:10.48550/arxiv.2605.09995","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2605.09995","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.2605.09995","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5035773495","display_name":"Jacob M. Springer","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Springer, Jacob Mitchell","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5135964368","display_name":"Madhu Advani","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Advani, Madhu","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5092441982","display_name":"Lukas Aichberger","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Aichberger, Lukas","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5076803200","display_name":"Arwen Bradley","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Bradley, Arwen","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5069983576","display_name":"Eran Malach","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Malach, Eran","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5011650393","display_name":"Omid Saremi","orcid":"https://orcid.org/0000-0001-6718-093X"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Saremi, Omid","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5130325089","display_name":"Sinead Williamson","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Williamson, Sinead","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5110716768","display_name":"Preetum Nakkiran","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Nakkiran, Preetum","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5049929304","display_name":"Etai Littwin","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Littwin, Etai","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5135963364","display_name":"Aditi Raghunathan","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Raghunathan, Aditi","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/T11307","display_name":"Domain Adaptation and Few-Shot Learning","score":0.27320000529289246,"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/T11307","display_name":"Domain Adaptation and Few-Shot Learning","score":0.27320000529289246,"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":0.26440000534057617,"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":0.08640000224113464,"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"}}],"keywords":[{"id":"https://openalex.org/keywords/annotation","display_name":"Annotation","score":0.6739000082015991},{"id":"https://openalex.org/keywords/inference","display_name":"Inference","score":0.6607000231742859},{"id":"https://openalex.org/keywords/mode","display_name":"Mode (computer interface)","score":0.5339000225067139},{"id":"https://openalex.org/keywords/diversity","display_name":"Diversity (politics)","score":0.5205000042915344},{"id":"https://openalex.org/keywords/semantics","display_name":"Semantics (computer science)","score":0.49230000376701355},{"id":"https://openalex.org/keywords/semantic-data-model","display_name":"Semantic data model","score":0.4074999988079071},{"id":"https://openalex.org/keywords/training-set","display_name":"Training set","score":0.3930000066757202}],"concepts":[{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.7148000001907349},{"id":"https://openalex.org/C2776321320","wikidata":"https://www.wikidata.org/wiki/Q857525","display_name":"Annotation","level":2,"score":0.6739000082015991},{"id":"https://openalex.org/C2776214188","wikidata":"https://www.wikidata.org/wiki/Q408386","display_name":"Inference","level":2,"score":0.6607000231742859},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.5716000199317932},{"id":"https://openalex.org/C48677424","wikidata":"https://www.wikidata.org/wiki/Q6888088","display_name":"Mode (computer interface)","level":2,"score":0.5339000225067139},{"id":"https://openalex.org/C2781316041","wikidata":"https://www.wikidata.org/wiki/Q1230584","display_name":"Diversity (politics)","level":2,"score":0.5205000042915344},{"id":"https://openalex.org/C184337299","wikidata":"https://www.wikidata.org/wiki/Q1437428","display_name":"Semantics (computer science)","level":2,"score":0.49230000376701355},{"id":"https://openalex.org/C90312973","wikidata":"https://www.wikidata.org/wiki/Q7449052","display_name":"Semantic data model","level":2,"score":0.4074999988079071},{"id":"https://openalex.org/C51632099","wikidata":"https://www.wikidata.org/wiki/Q3985153","display_name":"Training set","level":2,"score":0.3930000066757202},{"id":"https://openalex.org/C198531522","wikidata":"https://www.wikidata.org/wiki/Q485146","display_name":"Sample (material)","level":2,"score":0.3617999851703644},{"id":"https://openalex.org/C204321447","wikidata":"https://www.wikidata.org/wiki/Q30642","display_name":"Natural language processing","level":1,"score":0.35429999232292175},{"id":"https://openalex.org/C130318100","wikidata":"https://www.wikidata.org/wiki/Q2268914","display_name":"Semantic similarity","level":2,"score":0.3476000130176544},{"id":"https://openalex.org/C119857082","wikidata":"https://www.wikidata.org/wiki/Q2539","display_name":"Machine learning","level":1,"score":0.3472000062465668},{"id":"https://openalex.org/C136197465","wikidata":"https://www.wikidata.org/wiki/Q1729295","display_name":"Variety (cybernetics)","level":2,"score":0.3255999982357025},{"id":"https://openalex.org/C2777211547","wikidata":"https://www.wikidata.org/wiki/Q17141490","display_name":"Training (meteorology)","level":2,"score":0.31949999928474426},{"id":"https://openalex.org/C110121322","wikidata":"https://www.wikidata.org/wiki/Q865811","display_name":"Distribution (mathematics)","level":2,"score":0.31630000472068787},{"id":"https://openalex.org/C75917345","wikidata":"https://www.wikidata.org/wiki/Q2725298","display_name":"Sampling bias","level":3,"score":0.30550000071525574},{"id":"https://openalex.org/C149441793","wikidata":"https://www.wikidata.org/wiki/Q200726","display_name":"Probability distribution","level":2,"score":0.26510000228881836},{"id":"https://openalex.org/C124101348","wikidata":"https://www.wikidata.org/wiki/Q172491","display_name":"Data mining","level":1,"score":0.25220000743865967}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.48550/arxiv.2605.09995","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2605.09995","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.2605.09995","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2605.09995","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":[{"id":"https://metadata.un.org/sdg/4","score":0.5640069842338562,"display_name":"Quality Education"}],"awards":[],"funders":[],"has_content":{"grobid_xml":false,"pdf":false},"content_urls":null,"referenced_works_count":0,"referenced_works":[],"related_works":[],"abstract_inverted_index":{"Post-training":[0],"(via":[1],"supervised":[2],"fine-tuning)":[3],"improves":[4],"instruction-following,":[5],"but":[6],"often":[7],"induces":[8],"semantic":[9,38,75,117],"mode":[10],"collapse":[11,136],"by":[12],"biasing":[13],"models":[14,50,125,138],"toward":[15],"low-entropy":[16],"fine-tuning":[17],"data":[18],"at":[19,103],"the":[20,23,53,60,84],"expense":[21],"of":[22,56,63,87],"high-entropy":[24],"pretraining":[25,88],"distribution.":[26],"Crucially,":[27],"we":[28,41,69,91],"find":[29,123],"this":[30,37,93],"trade-off":[31],"worsens":[32],"with":[33,74,127,140,144],"scale.":[34,145],"To":[35],"close":[36],"diversity":[39,62,135],"gap,":[40],"propose":[42],"annotation-anchored":[43,128],"training,":[44],"a":[45,78],"principled":[46],"method":[47],"that":[48,82,124],"enables":[49],"to":[51,111],"adopt":[52],"preference-following":[54],"behaviors":[55],"post-training":[57],"without":[58],"sacrificing":[59],"inherent":[61],"pretraining.":[64],"Our":[65],"approach":[66],"is":[67],"simple:":[68],"pretrain":[70],"on":[71],"documents":[72],"paired":[73],"annotations,":[76],"inducing":[77],"rich":[79],"annotation":[80],"distribution":[81,94],"reflects":[83],"full":[85],"breadth":[86],"data,":[89],"and":[90,106,142],"preserve":[92],"during":[95],"post-training.":[96],"This":[97],"lets":[98],"us":[99],"sample":[100],"diverse":[101],"annotations":[102],"inference":[104],"time":[105],"use":[107],"them":[108],"as":[109],"anchors":[110],"guide":[112],"generation,":[113],"effectively":[114],"transferring":[115],"pretraining's":[116],"richness":[118],"into":[119],"post-trained":[120],"models.":[121],"We":[122],"trained":[126,139],"training":[129],"can":[130],"attain":[131],"$6":[132],"\\times$":[133],"less":[134],"than":[137],"SFT,":[141],"improve":[143]},"counts_by_year":[],"updated_date":"2026-07-01T06:00:48.157686","created_date":"2026-05-13T00:00:00"}