{"id":"https://openalex.org/W7161237900","doi":"https://doi.org/10.48550/arxiv.2605.14749","title":"Non-linear Interventions on Large Language Models","display_name":"Non-linear Interventions on Large Language Models","publication_year":2026,"publication_date":"2026-05-14","ids":{"openalex":"https://openalex.org/W7161237900","doi":"https://doi.org/10.48550/arxiv.2605.14749"},"language":"en","primary_location":{"id":"pmh:oai:arXiv.org:2605.14749","doi":null,"is_oa":true,"landing_page_url":"https://arxiv.org/abs/2605.14749","pdf_url":"https://arxiv.org/pdf/2605.14749","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","indexed_in":["arxiv","datacite"],"open_access":{"is_oa":true,"oa_status":"green","oa_url":"https://arxiv.org/pdf/2605.14749","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5135477920","display_name":"Sangwoo Kim","orcid":null,"observed_orcids":[]},"institutions":[],"countries":[],"is_corresponding":true,"raw_author_name":"Kim, Sangwoo","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]}],"institutions":[],"countries_distinct_count":0,"institutions_distinct_count":0,"corresponding_author_ids":["https://openalex.org/A5135477920"],"corresponding_institution_ids":[],"apc_list":null,"apc_paid":null,"fwci":null,"has_fulltext":true,"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/T12026","display_name":"Explainable Artificial Intelligence (XAI)","score":0.9717000126838684,"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":0.9717000126838684,"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":0.017100000753998756,"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":0.00279999990016222,"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/intervention","display_name":"intervention","score":0.6430000066757202},{"id":"https://openalex.org/keywords/large-language-models","display_name":"large language models","score":0.9890000224113464},{"id":"https://openalex.org/keywords/linear-representation-hypothesis","display_name":"linear representation hypothesis","score":0.8859999775886536},{"id":"https://openalex.org/keywords/internal-representations","display_name":"internal representations","score":0.6669999957084656},{"id":"https://openalex.org/keywords/implicit-features","display_name":"implicit features","score":0.6890000104904175},{"id":"https://openalex.org/keywords/steering","display_name":"steering","score":0.32100000977516174},{"id":"https://openalex.org/keywords/interpretability","display_name":"interpretability","score":0.6320000290870667}],"concepts":[{"id":"https://openalex.org/C27415008","wikidata":"https://www.wikidata.org/wiki/Q7256382","display_name":"Psychological intervention","level":2,"score":0.7024000287055969},{"id":"https://openalex.org/C2780665704","wikidata":"https://www.wikidata.org/wiki/Q959298","display_name":"Intervention (counseling)","level":2,"score":0.6855999827384949},{"id":"https://openalex.org/C2776359362","wikidata":"https://www.wikidata.org/wiki/Q2145286","display_name":"Representation (politics)","level":3,"score":0.6388999819755554},{"id":"https://openalex.org/C2776401178","wikidata":"https://www.wikidata.org/wiki/Q12050496","display_name":"Feature (linguistics)","level":2,"score":0.5945000052452087},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.5551000237464905},{"id":"https://openalex.org/C163175372","wikidata":"https://www.wikidata.org/wiki/Q3339222","display_name":"Linear model","level":2,"score":0.4431000053882599},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.42570000886917114},{"id":"https://openalex.org/C137293760","wikidata":"https://www.wikidata.org/wiki/Q3621696","display_name":"Language model","level":2,"score":0.3815000057220459},{"id":"https://openalex.org/C15744967","wikidata":"https://www.wikidata.org/wiki/Q9418","display_name":"Psychology","level":0,"score":0.36739999055862427},{"id":"https://openalex.org/C204321447","wikidata":"https://www.wikidata.org/wiki/Q30642","display_name":"Natural language processing","level":1,"score":0.3560999929904938},{"id":"https://openalex.org/C119857082","wikidata":"https://www.wikidata.org/wiki/Q2539","display_name":"Machine learning","level":1,"score":0.3504999876022339},{"id":"https://openalex.org/C168167062","wikidata":"https://www.wikidata.org/wiki/Q1117970","display_name":"Component (thermodynamics)","level":2,"score":0.31850001215934753},{"id":"https://openalex.org/C204323151","wikidata":"https://www.wikidata.org/wiki/Q905424","display_name":"Range (aeronautics)","level":2,"score":0.3160000145435333},{"id":"https://openalex.org/C180747234","wikidata":"https://www.wikidata.org/wiki/Q23373","display_name":"Cognitive psychology","level":1,"score":0.2800000011920929},{"id":"https://openalex.org/C2777212361","wikidata":"https://www.wikidata.org/wiki/Q5127848","display_name":"Class (philosophy)","level":2,"score":0.27469998598098755},{"id":"https://openalex.org/C195324797","wikidata":"https://www.wikidata.org/wiki/Q33742","display_name":"Natural language","level":2,"score":0.2565999925136566},{"id":"https://openalex.org/C59656382","wikidata":"https://www.wikidata.org/wiki/Q191536","display_name":"Conjunction (astronomy)","level":2,"score":0.2563999891281128},{"id":"https://openalex.org/C161301231","wikidata":"https://www.wikidata.org/wiki/Q3478658","display_name":"Knowledge representation and reasoning","level":2,"score":0.25290000438690186}],"mesh":[],"locations_count":2,"locations":[{"id":"pmh:oai:arXiv.org:2605.14749","doi":null,"is_oa":true,"landing_page_url":"https://arxiv.org/abs/2605.14749","pdf_url":"https://arxiv.org/pdf/2605.14749","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.2605.14749","doi":"https://doi.org/10.48550/arxiv.2605.14749","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2605.14749","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:2605.14749","doi":null,"is_oa":true,"landing_page_url":"https://arxiv.org/abs/2605.14749","pdf_url":"https://arxiv.org/pdf/2605.14749","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":[],"has_content":{"pdf":true,"grobid_xml":true},"content_urls":{"pdf":"https://content.openalex.org/works/W7161237900.pdf","grobid_xml":"https://content.openalex.org/works/W7161237900.grobid-xml"},"referenced_works_count":0,"referenced_works":[],"related_works":["https://openalex.org/W7134884294","https://openalex.org/W7128000853","https://openalex.org/W4407764896","https://openalex.org/W7163597073","https://openalex.org/W7122648810","https://openalex.org/W7164029060","https://openalex.org/W7161099702","https://openalex.org/W7125350820","https://openalex.org/W4406603780","https://openalex.org/W7124175571"],"abstract_inverted_index":{"Intervention":[0],"is":[1],"one":[2],"of":[3,16,53],"the":[4,13,32,90],"most":[5],"representative":[6],"and":[7],"widely":[8],"used":[9],"methods":[10,24],"for":[11],"understanding":[12],"internal":[14],"representations":[15],"large":[17],"language":[18],"models":[19],"(LLMs).":[20],"However,":[21],"existing":[22],"intervention":[23,54,70],"are":[25],"confined":[26],"to":[27,58],"linear":[28,95],"interventions":[29],"grounded":[30],"in":[31],"Linear":[33],"Representation":[34],"Hypothesis,":[35],"leaving":[36],"features":[37,73],"encoded":[38],"along":[39],"non-linear":[40,101],"manifolds":[41],"beyond":[42],"their":[43],"reach.":[44],"In":[45],"this":[46],"work,":[47],"we":[48],"introduce":[49],"a":[50,64,75,100],"general":[51],"formulation":[52],"that":[55,67],"extends":[56],"naturally":[57],"non-linearly":[59],"represented":[60],"features,":[61],"together":[62],"with":[63],"learning":[65],"procedure":[66],"further":[68],"enables":[69],"on":[71,83,99],"implicit":[72],"lacking":[74],"direct":[76],"output":[77],"signature.":[78],"We":[79],"validate":[80],"our":[81],"framework":[82],"refusal":[84],"bypass":[85],"steering,":[86],"where":[87],"it":[88],"steers":[89],"model":[91],"more":[92],"precisely":[93],"than":[94],"baselines":[96],"by":[97],"intervening":[98],"feature":[102],"governing":[103],"refusal.":[104]},"counts_by_year":[],"updated_date":"2026-11-04T10:54:05.112534","created_date":"2026-05-16T00:00:00"}