{"id":"https://openalex.org/W6891980433","doi":"https://doi.org/10.48550/arxiv.2506.18167","title":"Understanding Reasoning in Thinking Language Models via Steering Vectors","display_name":"Understanding Reasoning in Thinking Language Models via Steering Vectors","publication_year":2025,"publication_date":"2025-06-22","ids":{"openalex":"https://openalex.org/W6891980433","doi":"https://doi.org/10.48550/arxiv.2506.18167"},"language":"en","primary_location":{"id":"pmh:oai:arXiv.org:2506.18167","doi":null,"is_oa":true,"landing_page_url":"http://arxiv.org/abs/2506.18167","pdf_url":"https://arxiv.org/pdf/2506.18167","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":null,"license_id":null,"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/2506.18167","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5092958466","display_name":"Constantin Venhoff","orcid":null,"observed_orcids":[]},"institutions":[{"id":"https://openalex.org/I24354313","display_name":"Universidad de Buenos Aires","ror":"https://ror.org/0081fs513","country_code":"AR","type":"education","lineage":["https://openalex.org/I24354313"]}],"countries":["AR"],"is_corresponding":false,"raw_author_name":"Venhoff, Constantin","raw_affiliation_strings":["University of Buenos Aires Argentina"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"University of Buenos Aires Argentina","institution_ids":["https://openalex.org/I24354313"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5107619598","display_name":"Iv\u00e1n Arcuschin","orcid":"https://orcid.org/0000-0001-5805-8828","observed_orcids":["https://orcid.org/0000-0001-5805-8828"]},"institutions":[{"id":"https://openalex.org/I40120149","display_name":"University of Oxford","ror":"https://ror.org/052gg0110","country_code":"GB","type":"education","lineage":["https://openalex.org/I40120149"]}],"countries":["GB"],"is_corresponding":false,"raw_author_name":"Arcuschin, Iv\u00e1n","raw_affiliation_strings":["University of Oxford United Kingdom"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"University of Oxford United Kingdom","institution_ids":["https://openalex.org/I40120149"]}]},{"author_position":"middle","author":{"id":null,"display_name":"Torr, Philip","orcid":null,"observed_orcids":[]},"institutions":[{"id":"https://openalex.org/I40120149","display_name":"University of Oxford","ror":"https://ror.org/052gg0110","country_code":"GB","type":"education","lineage":["https://openalex.org/I40120149"]}],"countries":["GB"],"is_corresponding":false,"raw_author_name":"Torr, Philip","raw_affiliation_strings":["University of Oxford United Kingdom"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"University of Oxford United Kingdom","institution_ids":["https://openalex.org/I40120149"]}]},{"author_position":"middle","author":{"id":null,"display_name":"Conmy, Arthur","orcid":null,"observed_orcids":[]},"institutions":[{"id":"https://openalex.org/I40120149","display_name":"University of Oxford","ror":"https://ror.org/052gg0110","country_code":"GB","type":"education","lineage":["https://openalex.org/I40120149"]}],"countries":["GB"],"is_corresponding":false,"raw_author_name":"Conmy, Arthur","raw_affiliation_strings":["University of Oxford United Kingdom"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"University of Oxford United Kingdom","institution_ids":["https://openalex.org/I40120149"]}]},{"author_position":"last","author":{"id":"https://openalex.org/A5044489746","display_name":"Neel Nanda","orcid":null,"observed_orcids":[]},"institutions":[{"id":"https://openalex.org/I40120149","display_name":"University of Oxford","ror":"https://ror.org/052gg0110","country_code":"GB","type":"education","lineage":["https://openalex.org/I40120149"]}],"countries":["GB"],"is_corresponding":false,"raw_author_name":"Nanda, Neel","raw_affiliation_strings":["University of Oxford United Kingdom"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"University of Oxford United Kingdom","institution_ids":["https://openalex.org/I40120149"]}]}],"institutions":[],"countries_distinct_count":2,"institutions_distinct_count":2,"corresponding_author_ids":[],"corresponding_institution_ids":[],"apc_list":null,"apc_paid":null,"fwci":null,"has_fulltext":true,"cited_by_count":64,"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":true,"primary_topic":{"id":"https://openalex.org/T12026","display_name":"Explainable Artificial Intelligence (XAI)","score":0.7534000277519226,"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/T12026","display_name":"Explainable Artificial Intelligence (XAI)","score":0.7534000277519226,"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":0.17759999632835388,"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":0.02850000001490116,"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/steering-vectors","display_name":"steering vectors","score":0.8790000081062317},{"id":"https://openalex.org/keywords/reasoning-control","display_name":"reasoning control","score":0.6439999938011169},{"id":"https://openalex.org/keywords/activation-space","display_name":"activation space","score":0.5049999952316284},{"id":"https://openalex.org/keywords/uncertainty-expression","display_name":"uncertainty expression","score":0.9229999780654907},{"id":"https://openalex.org/keywords/hypothesis-validation","display_name":"hypothesis verification","score":0.7110000252723694},{"id":"https://openalex.org/keywords/interpretability","display_name":"interpretability","score":0.7310000061988831},{"id":"https://openalex.org/keywords/large-language-models","display_name":"large language models","score":0.628000020980835}],"concepts":[{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.6420000195503235},{"id":"https://openalex.org/C2775924081","wikidata":"https://www.wikidata.org/wiki/Q55608371","display_name":"Control (management)","level":2,"score":0.47130000591278076},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.4499000012874603},{"id":"https://openalex.org/C86827895","wikidata":"https://www.wikidata.org/wiki/Q7098582","display_name":"Opportunistic reasoning","level":4,"score":0.4404999911785126},{"id":"https://openalex.org/C89288958","wikidata":"https://www.wikidata.org/wiki/Q7301504","display_name":"Reasoning system","level":2,"score":0.43880000710487366},{"id":"https://openalex.org/C97364631","wikidata":"https://www.wikidata.org/wiki/Q484284","display_name":"Deductive reasoning","level":2,"score":0.4277999997138977},{"id":"https://openalex.org/C183521366","wikidata":"https://www.wikidata.org/wiki/Q7256422","display_name":"Psychology of reasoning","level":4,"score":0.427700012922287},{"id":"https://openalex.org/C195344581","wikidata":"https://www.wikidata.org/wiki/Q2555318","display_name":"Automated reasoning","level":2,"score":0.41110000014305115},{"id":"https://openalex.org/C155911833","wikidata":"https://www.wikidata.org/wiki/Q3817354","display_name":"Spatial intelligence","level":2,"score":0.4104999899864197},{"id":"https://openalex.org/C103057564","wikidata":"https://www.wikidata.org/wiki/Q4751139","display_name":"Analytic reasoning","level":3,"score":0.4090000092983246},{"id":"https://openalex.org/C156884757","wikidata":"https://www.wikidata.org/wiki/Q798554","display_name":"Backtracking","level":2,"score":0.37139999866485596},{"id":"https://openalex.org/C2778572836","wikidata":"https://www.wikidata.org/wiki/Q380933","display_name":"Space (punctuation)","level":2,"score":0.367900013923645},{"id":"https://openalex.org/C18762648","wikidata":"https://www.wikidata.org/wiki/Q42213","display_name":"Work (physics)","level":2,"score":0.36550000309944153},{"id":"https://openalex.org/C159032336","wikidata":"https://www.wikidata.org/wiki/Q2488768","display_name":"Non-monotonic logic","level":2,"score":0.350600004196167},{"id":"https://openalex.org/C188147891","wikidata":"https://www.wikidata.org/wiki/Q147638","display_name":"Cognitive science","level":1,"score":0.34790000319480896},{"id":"https://openalex.org/C20162079","wikidata":"https://www.wikidata.org/wiki/Q1151406","display_name":"Case-based reasoning","level":2,"score":0.33230000734329224},{"id":"https://openalex.org/C83725634","wikidata":"https://www.wikidata.org/wiki/Q7268699","display_name":"Qualitative reasoning","level":2,"score":0.3301999866962433},{"id":"https://openalex.org/C37335422","wikidata":"https://www.wikidata.org/wiki/Q6888134","display_name":"Model-based reasoning","level":3,"score":0.3133000135421753},{"id":"https://openalex.org/C187179951","wikidata":"https://www.wikidata.org/wiki/Q7784616","display_name":"Thinking processes","level":3,"score":0.310699999332428},{"id":"https://openalex.org/C115086926","wikidata":"https://www.wikidata.org/wiki/Q17004651","display_name":"Causal reasoning","level":3,"score":0.3100000023841858},{"id":"https://openalex.org/C36964233","wikidata":"https://www.wikidata.org/wiki/Q7920942","display_name":"Verbal reasoning","level":3,"score":0.3077999949455261},{"id":"https://openalex.org/C137293760","wikidata":"https://www.wikidata.org/wiki/Q3621696","display_name":"Language model","level":2,"score":0.30390000343322754},{"id":"https://openalex.org/C43971567","wikidata":"https://www.wikidata.org/wiki/Q3142865","display_name":"Logical reasoning","level":2,"score":0.2858999967575073},{"id":"https://openalex.org/C2777508537","wikidata":"https://www.wikidata.org/wiki/Q7936620","display_name":"Visual reasoning","level":2,"score":0.2824999988079071},{"id":"https://openalex.org/C107457646","wikidata":"https://www.wikidata.org/wiki/Q207434","display_name":"Human\u2013computer interaction","level":1,"score":0.2757999897003174},{"id":"https://openalex.org/C184337299","wikidata":"https://www.wikidata.org/wiki/Q1437428","display_name":"Semantics (computer science)","level":2,"score":0.27160000801086426},{"id":"https://openalex.org/C107848011","wikidata":"https://www.wikidata.org/wiki/Q4680756","display_name":"Adaptive reasoning","level":4,"score":0.2678999900817871},{"id":"https://openalex.org/C162100613","wikidata":"https://www.wikidata.org/wiki/Q944461","display_name":"Lateral thinking","level":4,"score":0.25920000672340393},{"id":"https://openalex.org/C98045186","wikidata":"https://www.wikidata.org/wiki/Q205663","display_name":"Process (computing)","level":2,"score":0.2498999983072281},{"id":"https://openalex.org/C28427503","wikidata":"https://www.wikidata.org/wiki/Q13580300","display_name":"Internal model","level":3,"score":0.24469999969005585},{"id":"https://openalex.org/C169900460","wikidata":"https://www.wikidata.org/wiki/Q2200417","display_name":"Cognition","level":2,"score":0.23929999768733978},{"id":"https://openalex.org/C108650721","wikidata":"https://www.wikidata.org/wiki/Q1783253","display_name":"Counterfactual thinking","level":2,"score":0.2337000072002411},{"id":"https://openalex.org/C195324797","wikidata":"https://www.wikidata.org/wiki/Q33742","display_name":"Natural language","level":2,"score":0.23029999434947968},{"id":"https://openalex.org/C161167727","wikidata":"https://www.wikidata.org/wiki/Q6111847","display_name":"Vertical thinking","level":5,"score":0.22759999334812164},{"id":"https://openalex.org/C175154964","wikidata":"https://www.wikidata.org/wiki/Q380077","display_name":"Task analysis","level":3,"score":0.22139999270439148},{"id":"https://openalex.org/C108867877","wikidata":"https://www.wikidata.org/wiki/Q5186773","display_name":"Critical systems thinking","level":3,"score":0.21719999611377716},{"id":"https://openalex.org/C2994050667","wikidata":"https://www.wikidata.org/wiki/Q673661","display_name":"Abstract reasoning","level":3,"score":0.21299999952316284},{"id":"https://openalex.org/C2780791683","wikidata":"https://www.wikidata.org/wiki/Q846785","display_name":"Action (physics)","level":2,"score":0.2126999944448471},{"id":"https://openalex.org/C2779439875","wikidata":"https://www.wikidata.org/wiki/Q1078276","display_name":"Natural language understanding","level":3,"score":0.21070000529289246},{"id":"https://openalex.org/C11671645","wikidata":"https://www.wikidata.org/wiki/Q5054567","display_name":"Causal model","level":2,"score":0.19760000705718994}],"mesh":[],"locations_count":2,"locations":[{"id":"pmh:oai:arXiv.org:2506.18167","doi":null,"is_oa":true,"landing_page_url":"http://arxiv.org/abs/2506.18167","pdf_url":"https://arxiv.org/pdf/2506.18167","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":null,"license_id":null,"version":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"text"},{"id":"doi:10.48550/arxiv.2506.18167","doi":"https://doi.org/10.48550/arxiv.2506.18167","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2506.18167","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":null,"license_id":null,"version":null,"is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"Preprint"}],"best_oa_location":{"id":"pmh:oai:arXiv.org:2506.18167","doi":null,"is_oa":true,"landing_page_url":"http://arxiv.org/abs/2506.18167","pdf_url":"https://arxiv.org/pdf/2506.18167","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":null,"license_id":null,"version":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"text"},"sustainable_development_goals":[],"study_designs":[],"awards":[],"funders":[{"id":"https://openalex.org/F4320313434","display_name":"Institute for Catastrophic Loss Reduction","ror":"https://ror.org/03prrhh04"}],"has_content":{"grobid_xml":true,"pdf":true},"content_urls":{"pdf":"https://content.openalex.org/works/W6891980433.pdf","grobid_xml":"https://content.openalex.org/works/W6891980433.grobid-xml"},"referenced_works_count":8,"referenced_works":["https://openalex.org/W4285149215","https://openalex.org/W4386081218","https://openalex.org/W4387355345","https://openalex.org/W4392340654","https://openalex.org/W4402671231","https://openalex.org/W4415800641","https://openalex.org/W7133235880","https://openalex.org/W7196987534"],"related_works":["https://openalex.org/W4417057051","https://openalex.org/W4416035889","https://openalex.org/W7148837110","https://openalex.org/W6948163397","https://openalex.org/W7119557443","https://openalex.org/W7163363052","https://openalex.org/W4414137972","https://openalex.org/W4416036504","https://openalex.org/W7118028454","https://openalex.org/W7127908594"],"abstract_inverted_index":{"Recent":[0],"advances":[1],"in":[2,53,86,99,148,151],"large":[3],"language":[4,14],"models":[5,15,27,150],"(LLMs)":[6],"have":[7],"led":[8],"to":[9,121,134],"the":[10,100,126],"development":[11],"of":[12,125],"thinking":[13,44,74,149],"that":[16,91],"generate":[17],"extensive":[18],"internal":[19],"reasoning":[20,33,51,70,87,128,146],"chains":[21],"before":[22],"producing":[23],"responses.":[24],"While":[25],"these":[26,92,115],"achieve":[28],"improved":[29],"performance,":[30],"controlling":[31],"their":[32],"processes":[34,147],"remains":[35],"challenging.":[36],"This":[37],"work":[38],"presents":[39],"a":[40,57,119,152],"steering":[41,109,145,160],"approach":[42,140],"for":[43,81,144],"LLMs":[45],"by":[46,73,96],"analyzing":[47],"and":[48,84,104,113,154],"manipulating":[49],"specific":[50,123],"behaviors":[52,71,93],"DeepSeek-R1-Distill":[54,164],"models.":[55],"Through":[56],"systematic":[58],"experiment":[59],"on":[60],"500":[61],"tasks":[62],"across":[63,169],"10":[64],"diverse":[65],"categories,":[66],"we":[67,117],"identify":[68],"several":[69],"exhibited":[72],"models,":[75,165],"including":[76],"expressing":[77],"uncertainty,":[78],"generating":[79],"examples":[80],"hypothesis":[82],"validation,":[83],"backtracking":[85],"chains.":[88],"We":[89,157],"demonstrate":[90],"are":[94],"mediated":[95],"linear":[97],"directions":[98],"model's":[101,127],"activation":[102],"space":[103],"can":[105],"be":[106],"controlled":[107,153],"using":[108,162],"vectors.":[110],"By":[111],"extracting":[112],"applying":[114],"vectors,":[116],"provide":[118],"method":[120,161],"modulate":[122],"aspects":[124],"process,":[129],"such":[130],"as":[131],"its":[132],"tendency":[133],"backtrack":[135],"or":[136],"express":[137],"uncertainty.":[138],"Our":[139],"offers":[141],"practical":[142],"tools":[143],"interpretable":[155],"manner.":[156],"validate":[158],"our":[159],"three":[163],"demonstrating":[166],"consistent":[167],"control":[168],"different":[170],"model":[171],"architectures.":[172]},"counts_by_year":[{"year":2026,"cited_by_count":49},{"year":2025,"cited_by_count":15}],"updated_date":"2026-11-11T12:16:26.678958","created_date":"2025-10-10T00:00:00"}