{"id":"https://openalex.org/W7147065729","doi":"https://doi.org/10.48550/arxiv.2603.30036","title":"Aligned, Orthogonal or In-conflict: When can we safely optimize Chain-of-Thought?","display_name":"Aligned, Orthogonal or In-conflict: When can we safely optimize Chain-of-Thought?","publication_year":2026,"publication_date":"2026-03-31","ids":{"openalex":"https://openalex.org/W7147065729","doi":"https://doi.org/10.48550/arxiv.2603.30036"},"language":null,"primary_location":{"id":"doi:10.48550/arxiv.2603.30036","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2603.30036","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":"article"},"type":"preprint","indexed_in":["datacite"],"open_access":{"is_oa":true,"oa_status":"green","oa_url":"https://doi.org/10.48550/arxiv.2603.30036","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5132688683","display_name":"Max Kaufmann","orcid":null},"institutions":[],"countries":[],"is_corresponding":true,"raw_author_name":"Kaufmann, Max","raw_affiliation_strings":[],"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5132573890","display_name":"David Lindner","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Lindner, David","raw_affiliation_strings":[],"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5102706612","display_name":"R. Zimmermann","orcid":"https://orcid.org/0000-0001-8790-1618"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Zimmermann, Roland S.","raw_affiliation_strings":[],"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5132560974","display_name":"and Rohin Shah","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Shah, and Rohin","raw_affiliation_strings":[],"affiliations":[]}],"institutions":[],"countries_distinct_count":0,"institutions_distinct_count":4,"corresponding_author_ids":["https://openalex.org/A5132688683"],"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/T12127","display_name":"Software System Performance and Reliability","score":0.17520000040531158,"subfield":{"id":"https://openalex.org/subfields/1705","display_name":"Computer Networks and Communications"},"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/T12127","display_name":"Software System Performance and Reliability","score":0.17520000040531158,"subfield":{"id":"https://openalex.org/subfields/1705","display_name":"Computer Networks and Communications"},"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":0.10260000079870224,"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/T10639","display_name":"Advanced Software Engineering Methodologies","score":0.07419999688863754,"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/set","display_name":"Set (abstract data type)","score":0.6371999979019165},{"id":"https://openalex.org/keywords/term","display_name":"Term (time)","score":0.5199999809265137},{"id":"https://openalex.org/keywords/training","display_name":"Training (meteorology)","score":0.4512999951839447},{"id":"https://openalex.org/keywords/training-set","display_name":"Training set","score":0.42980000376701355},{"id":"https://openalex.org/keywords/component","display_name":"Component (thermodynamics)","score":0.37529999017715454},{"id":"https://openalex.org/keywords/control","display_name":"Control (management)","score":0.3522000014781952}],"concepts":[{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.7200999855995178},{"id":"https://openalex.org/C177264268","wikidata":"https://www.wikidata.org/wiki/Q1514741","display_name":"Set (abstract data type)","level":2,"score":0.6371999979019165},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.574999988079071},{"id":"https://openalex.org/C61797465","wikidata":"https://www.wikidata.org/wiki/Q1188986","display_name":"Term (time)","level":2,"score":0.5199999809265137},{"id":"https://openalex.org/C119857082","wikidata":"https://www.wikidata.org/wiki/Q2539","display_name":"Machine learning","level":1,"score":0.461899995803833},{"id":"https://openalex.org/C2777211547","wikidata":"https://www.wikidata.org/wiki/Q17141490","display_name":"Training (meteorology)","level":2,"score":0.4512999951839447},{"id":"https://openalex.org/C51632099","wikidata":"https://www.wikidata.org/wiki/Q3985153","display_name":"Training set","level":2,"score":0.42980000376701355},{"id":"https://openalex.org/C168167062","wikidata":"https://www.wikidata.org/wiki/Q1117970","display_name":"Component (thermodynamics)","level":2,"score":0.37529999017715454},{"id":"https://openalex.org/C2775924081","wikidata":"https://www.wikidata.org/wiki/Q55608371","display_name":"Control (management)","level":2,"score":0.3522000014781952},{"id":"https://openalex.org/C26517878","wikidata":"https://www.wikidata.org/wiki/Q228039","display_name":"Key (lock)","level":2,"score":0.349700003862381},{"id":"https://openalex.org/C2776035688","wikidata":"https://www.wikidata.org/wiki/Q1606558","display_name":"Affect (linguistics)","level":2,"score":0.34049999713897705},{"id":"https://openalex.org/C2776401178","wikidata":"https://www.wikidata.org/wiki/Q12050496","display_name":"Feature (linguistics)","level":2,"score":0.29660001397132874},{"id":"https://openalex.org/C97541855","wikidata":"https://www.wikidata.org/wiki/Q830687","display_name":"Reinforcement learning","level":2,"score":0.26840001344680786},{"id":"https://openalex.org/C13606891","wikidata":"https://www.wikidata.org/wiki/Q2623243","display_name":"Conceptual model","level":2,"score":0.2565000057220459},{"id":"https://openalex.org/C124101348","wikidata":"https://www.wikidata.org/wiki/Q172491","display_name":"Data mining","level":1,"score":0.2554999887943268}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.48550/arxiv.2603.30036","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2603.30036","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":"article"}],"best_oa_location":{"id":"doi:10.48550/arxiv.2603.30036","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2603.30036","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":"article"},"sustainable_development_goals":[],"awards":[],"funders":[],"has_content":{"pdf":false,"grobid_xml":false},"content_urls":null,"referenced_works_count":0,"referenced_works":[],"related_works":[],"abstract_inverted_index":{"Chain-of-Thought":[0],"(CoT)":[1],"monitoring,":[2],"in":[3],"which":[4,26],"automated":[5],"systems":[6],"monitor":[7],"the":[8,23,33,36,39,50,84,101],"CoT":[9,29,40,165,177],"of":[10,38,57,152],"an":[11,80],"LLM,":[12],"is":[13,185],"a":[14,27,65,150],"promising":[15],"approach":[16],"for":[17,47,68],"effectively":[18],"overseeing":[19],"AI":[20],"systems.":[21],"However,":[22],"extent":[24],"to":[25,53,107,148],"model's":[28],"helps":[30],"us":[31,106],"oversee":[32],"model":[34,51,76],"-":[35,41],"monitorability":[37,178],"can":[42],"be":[43],"affected":[44],"by":[45,49],"training,":[46],"instance":[48],"learning":[52],"hide":[54],"important":[55],"features":[56],"its":[58],"reasoning.":[59],"We":[60,75,119,167],"propose":[61],"and":[62,71,96,135,160,179],"empirically":[63],"validate":[64,142],"conceptual":[66],"framework":[67,104],"predicting":[69],"when":[70],"why":[72],"this":[73],"occurs.":[74],"LLM":[77],"post-training":[78],"as":[79,112],"RL":[81,153],"environment":[82],"where":[83],"reward":[85,174,183],"decomposes":[86],"into":[87],"two":[88,110],"terms:":[89],"one":[90],"term":[91,98],"depending":[92,99],"on":[93,100],"final":[94],"outputs":[95],"another":[97],"CoT.":[102],"Our":[103],"allows":[105],"classify":[108,149],"these":[109],"terms":[111,125,130,137,175,184],"\"aligned\",":[113],"\"orthogonal\",":[114],"or":[115],"\"in-conflict\"":[116,173],"before":[117],"training.":[118],"predict":[120],"that":[121,169],"training":[122,163,171],"with":[123,172],"in-conflict":[124,182],"will":[126,131,138],"reduce":[127],"monitorability,":[128],"orthogonal":[129],"not":[132],"affect":[133],"it,":[134],"aligned":[136],"improve":[139],"it.":[140],"To":[141],"our":[143],"framework,":[144],"we":[145],"use":[146],"it":[147],"set":[151],"environments,":[154,159],"train":[155],"LLMs":[156],"within":[157],"those":[158],"evaluate":[161],"how":[162],"affects":[164],"monitorability.":[166],"find":[168],"(1)":[170],"reduces":[176],"(2)":[180],"optimizing":[181],"difficult.":[186]},"counts_by_year":[],"updated_date":"2026-04-02T13:53:19.096889","created_date":"2026-04-02T00:00:00"}