{"id":"https://openalex.org/W7164344040","doi":"https://doi.org/10.48550/arxiv.2606.11599","title":"When is Your LLM Steerable?","display_name":"When is Your LLM Steerable?","publication_year":2026,"publication_date":"2026-06-10","ids":{"openalex":"https://openalex.org/W7164344040","doi":"https://doi.org/10.48550/arxiv.2606.11599"},"language":null,"primary_location":{"id":"doi:10.48550/arxiv.2606.11599","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2606.11599","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":null,"license_id":null,"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.2606.11599","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5138409266","display_name":"Chenrui Fan","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Fan, Chenrui","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5124822953","display_name":"Yize Cheng","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Cheng, Yize","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5138403406","display_name":"Ming Li (91180)","orcid":"https://orcid.org/0009-0000-8774-2514"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Li, Ming","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5025450606","display_name":"Soheil Feizi","orcid":"https://orcid.org/0000-0003-0944-8242"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Feizi, Soheil","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5138454071","display_name":"Tianyi Zhou","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Zhou, Tianyi","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":0.2402999997138977,"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":0.2402999997138977,"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":0.17589999735355377,"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/T11948","display_name":"Machine Learning in Materials Science","score":0.05180000141263008,"subfield":{"id":"https://openalex.org/subfields/2505","display_name":"Materials Chemistry"},"field":{"id":"https://openalex.org/fields/25","display_name":"Materials Science"},"domain":{"id":"https://openalex.org/domains/3","display_name":"Physical Sciences"}}],"keywords":[{"id":"https://openalex.org/keywords/security-token","display_name":"Security token","score":0.5997999906539917},{"id":"https://openalex.org/keywords/decoding-methods","display_name":"Decoding methods","score":0.5949000120162964},{"id":"https://openalex.org/keywords/encode","display_name":"ENCODE","score":0.5738000273704529},{"id":"https://openalex.org/keywords/leverage","display_name":"Leverage (statistics)","score":0.553600013256073},{"id":"https://openalex.org/keywords/inference","display_name":"Inference","score":0.5375000238418579},{"id":"https://openalex.org/keywords/classifier","display_name":"Classifier (UML)","score":0.5353000164031982},{"id":"https://openalex.org/keywords/testbed","display_name":"Testbed","score":0.4438999891281128},{"id":"https://openalex.org/keywords/boosting","display_name":"Boosting (machine learning)","score":0.32760000228881836}],"concepts":[{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.6589999794960022},{"id":"https://openalex.org/C48145219","wikidata":"https://www.wikidata.org/wiki/Q1335365","display_name":"Security token","level":2,"score":0.5997999906539917},{"id":"https://openalex.org/C57273362","wikidata":"https://www.wikidata.org/wiki/Q576722","display_name":"Decoding methods","level":2,"score":0.5949000120162964},{"id":"https://openalex.org/C66746571","wikidata":"https://www.wikidata.org/wiki/Q1134833","display_name":"ENCODE","level":3,"score":0.5738000273704529},{"id":"https://openalex.org/C153083717","wikidata":"https://www.wikidata.org/wiki/Q6535263","display_name":"Leverage (statistics)","level":2,"score":0.553600013256073},{"id":"https://openalex.org/C2776214188","wikidata":"https://www.wikidata.org/wiki/Q408386","display_name":"Inference","level":2,"score":0.5375000238418579},{"id":"https://openalex.org/C95623464","wikidata":"https://www.wikidata.org/wiki/Q1096149","display_name":"Classifier (UML)","level":2,"score":0.5353000164031982},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.48989999294281006},{"id":"https://openalex.org/C31395832","wikidata":"https://www.wikidata.org/wiki/Q1318674","display_name":"Testbed","level":2,"score":0.4438999891281128},{"id":"https://openalex.org/C119857082","wikidata":"https://www.wikidata.org/wiki/Q2539","display_name":"Machine learning","level":1,"score":0.4059999883174896},{"id":"https://openalex.org/C46686674","wikidata":"https://www.wikidata.org/wiki/Q466303","display_name":"Boosting (machine learning)","level":2,"score":0.32760000228881836},{"id":"https://openalex.org/C2778858076","wikidata":"https://www.wikidata.org/wiki/Q5249539","display_name":"Decodes","level":3,"score":0.30959999561309814},{"id":"https://openalex.org/C159877910","wikidata":"https://www.wikidata.org/wiki/Q2202883","display_name":"Autoregressive model","level":2,"score":0.3075999915599823},{"id":"https://openalex.org/C90509273","wikidata":"https://www.wikidata.org/wiki/Q11012","display_name":"Robot","level":2,"score":0.3037000000476837},{"id":"https://openalex.org/C26517878","wikidata":"https://www.wikidata.org/wiki/Q228039","display_name":"Key (lock)","level":2,"score":0.29420000314712524},{"id":"https://openalex.org/C2779803651","wikidata":"https://www.wikidata.org/wiki/Q5282088","display_name":"Discriminator","level":3,"score":0.29269999265670776},{"id":"https://openalex.org/C179518139","wikidata":"https://www.wikidata.org/wiki/Q5140297","display_name":"Coding (social sciences)","level":2,"score":0.2888999879360199},{"id":"https://openalex.org/C187691185","wikidata":"https://www.wikidata.org/wiki/Q2020720","display_name":"Grid","level":2,"score":0.28790000081062317},{"id":"https://openalex.org/C70153297","wikidata":"https://www.wikidata.org/wiki/Q5591907","display_name":"Gradient boosting","level":3,"score":0.28780001401901245}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.48550/arxiv.2606.11599","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2606.11599","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":null,"license_id":null,"version":null,"is_accepted":false,"is_published":null,"raw_source_name":null,"raw_type":"Preprint"}],"best_oa_location":{"id":"doi:10.48550/arxiv.2606.11599","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2606.11599","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":null,"license_id":null,"version":null,"is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"Preprint"},"sustainable_development_goals":[],"awards":[],"funders":[],"has_content":{"grobid_xml":false,"pdf":false},"content_urls":null,"referenced_works_count":0,"referenced_works":[],"related_works":[],"abstract_inverted_index":{"Activation":[0],"steering":[1,28,37,88,109,132,208,219],"offers":[2],"a":[3,84,98,163,226],"lightweight":[4],"approach":[5],"to":[6,81,86,173],"control":[7],"language":[8],"models'":[9],"behavior":[10],"at":[11,65],"inference":[12],"time,":[13],"but":[14],"whether":[15,55,175],"it":[16],"succeeds":[17],"or":[18,181],"fails":[19],"heavily":[20],"depends":[21],"on":[22,170,194],"the":[23,31,61,66,69,75,117],"prompt,":[24],"concept,":[25],"model,":[26],"and":[27,33,43,79,130,135,150],"configuration.":[29],"Finding":[30],"regime":[32],"boundaries":[34],"of":[35,46,68,229],"successful":[36],"typically":[38],"requires":[39],"expensive":[40],"grid":[41],"searches":[42],"post-hoc":[44],"evaluation":[45],"full":[47,185],"autoregressive":[48],"rollouts.":[49],"In":[50],"this":[51,92,113,213],"work,":[52],"we":[53,94,115],"investigate":[54],"steerability":[56,158,214],"can":[57],"be":[58],"predicted":[59],"from":[60],"model's":[62,118],"internal":[63],"states":[64,128,201],"beginning":[67],"generation":[70],"process,":[71],"e.g.,":[72],"after":[73,131],"generating":[74],"first":[76,95],"few":[77],"tokens,":[78],"how":[80,144],"leverage":[82,212],"such":[83],"predictor":[85,188,215],"improve":[87],"success":[89],"rate.":[90],"To":[91],"end,":[93],"introduce":[96],"ASTEER,":[97],"testbed":[99],"including":[100],"1.4M":[101],"steered":[102],"generations,":[103],"spanning":[104],"150":[105],"concepts":[106],"with":[107,225],"each":[108],"success/failure":[110],"labeled.":[111],"Leveraging":[112],"testbed,":[114],"analyze":[116],"early":[119,199],"decoding":[120,137,230],"dynamics":[121],"by":[122],"extracting":[123],"features":[124,140,172],"that":[125,198],"compare":[126],"hidden":[127,200],"before":[129],"across":[133],"layers":[134,149],"initial":[136],"steps.":[138],"These":[139],"help":[141],"us":[142],"understand":[143],"steering's":[145],"effects":[146],"propagate":[147],"along":[148],"token":[151],"positions,":[152],"which":[153],"provide":[154],"key":[155],"information":[156,205],"for":[157,218],"prediction.":[159],"We":[160,210],"then":[161],"train":[162],"Gradient":[164],"Boosting":[165],"Decision":[166],"Trees":[167],"(GBDT)":[168],"classifier":[169],"these":[171],"predict":[174],"an":[176],"intervention":[177],"will":[178],"under-steer,":[179],"succeed,":[180],"over-steer":[182],"without":[183],"requiring":[184],"rollout.":[186],"Our":[187],"achieves":[189],"around":[190],"0.7":[191],"macro-F1":[192],"score":[193],"unseen":[195],"concepts,":[196],"demonstrating":[197],"encode":[202],"substantial,":[203],"structured":[204],"about":[206],"eventual":[207],"efficacy.":[209],"further":[211],"as":[216],"guidance":[217],"strength":[220],"searching,":[221],"achieving":[222],"near-optimal":[223],"performance":[224],"small":[227],"fraction":[228],"cost.":[231]},"counts_by_year":[],"updated_date":"2026-07-01T06:00:48.157686","created_date":"2026-06-12T00:00:00"}