{"id":"https://openalex.org/W4392736799","doi":"https://doi.org/10.48550/arxiv.2403.05767","title":"Extending Activation Steering to Broad Skills and Multiple Behaviours","display_name":"Extending Activation Steering to Broad Skills and Multiple Behaviours","publication_year":2024,"publication_date":"2024-03-09","ids":{"openalex":"https://openalex.org/W4392736799","doi":"https://doi.org/10.48550/arxiv.2403.05767"},"language":"en","primary_location":{"id":"pmh:oai:arXiv.org:2403.05767","doi":null,"is_oa":true,"landing_page_url":"http://arxiv.org/abs/2403.05767","pdf_url":"https://arxiv.org/pdf/2403.05767","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":"cc-by","license_id":"https://openalex.org/licenses/cc-by","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/2403.05767","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5092401144","display_name":"Teun van der Weij","orcid":null,"observed_orcids":[]},"institutions":[{"id":"https://openalex.org/I193662353","display_name":"Utrecht University","ror":"https://ror.org/04pp8hn57","country_code":"NL","type":"education","lineage":["https://openalex.org/I193662353"]}],"countries":["NL"],"is_corresponding":false,"raw_author_name":"van der Weij, Teun","raw_affiliation_strings":["Utrecht University"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Utrecht University","institution_ids":["https://openalex.org/I193662353"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5047065550","display_name":"Massimo Poesio","orcid":"https://orcid.org/0000-0001-8469-2072","observed_orcids":["https://orcid.org/0000-0001-8469-2072"]},"institutions":[{"id":"https://openalex.org/I193662353","display_name":"Utrecht University","ror":"https://ror.org/04pp8hn57","country_code":"NL","type":"education","lineage":["https://openalex.org/I193662353"]}],"countries":["NL"],"is_corresponding":false,"raw_author_name":"Poesio, Massimo","raw_affiliation_strings":["Utrecht University"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Utrecht University","institution_ids":["https://openalex.org/I193662353"]}]},{"author_position":"last","author":{"id":"https://openalex.org/A5062598891","display_name":"Nandi Schoots","orcid":null,"observed_orcids":[]},"institutions":[{"id":"https://openalex.org/I183935753","display_name":"King's College London","ror":"https://ror.org/0220mzb33","country_code":"GB","type":"education","lineage":["https://openalex.org/I124357947","https://openalex.org/I183935753"]}],"countries":["GB"],"is_corresponding":false,"raw_author_name":"Schoots, Nandi","raw_affiliation_strings":["King's College London"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"King's College London","institution_ids":["https://openalex.org/I183935753"]}]}],"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":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/T11689","display_name":"Adversarial Robustness in Machine Learning","score":0.7939000129699707,"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/T11689","display_name":"Adversarial Robustness in Machine Learning","score":0.7939000129699707,"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/T10883","display_name":"Ethics and Social Impacts of AI","score":0.13830000162124634,"subfield":{"id":"https://openalex.org/subfields/3311","display_name":"Safety Research"},"field":{"id":"https://openalex.org/fields/33","display_name":"Social Sciences"},"domain":{"id":"https://openalex.org/domains/2","display_name":"Social Sciences"}},{"id":"https://openalex.org/T10028","display_name":"Topic Modeling","score":0.018200000748038292,"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/activation-steering","display_name":"activation steering","score":0.996999979019165},{"id":"https://openalex.org/keywords/large-language-models","display_name":"large language models","score":0.9440000057220459},{"id":"https://openalex.org/keywords/myopia","display_name":"myopia","score":0.6740000247955322},{"id":"https://openalex.org/keywords/steering-vectors","display_name":"steering vectors","score":0.5400000214576721}],"concepts":[{"id":"https://openalex.org/C15744967","wikidata":"https://www.wikidata.org/wiki/Q9418","display_name":"Psychology","level":0,"score":0.4294569492340088}],"mesh":[],"locations_count":2,"locations":[{"id":"pmh:oai:arXiv.org:2403.05767","doi":null,"is_oa":true,"landing_page_url":"http://arxiv.org/abs/2403.05767","pdf_url":"https://arxiv.org/pdf/2403.05767","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":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"text"},{"id":"doi:10.48550/arxiv.2403.05767","doi":"https://doi.org/10.48550/arxiv.2403.05767","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2403.05767","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":"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"}],"best_oa_location":{"id":"pmh:oai:arXiv.org:2403.05767","doi":null,"is_oa":true,"landing_page_url":"http://arxiv.org/abs/2403.05767","pdf_url":"https://arxiv.org/pdf/2403.05767","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":"cc-by","license_id":"https://openalex.org/licenses/cc-by","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/F4320321010","display_name":"Universiteit Utrecht","ror":"https://ror.org/04pp8hn57"}],"has_content":{"pdf":true,"grobid_xml":true},"content_urls":{"pdf":"https://content.openalex.org/works/W4392736799.pdf","grobid_xml":"https://content.openalex.org/works/W4392736799.grobid-xml"},"referenced_works_count":0,"referenced_works":[],"related_works":["https://openalex.org/W7106653175","https://openalex.org/W7220918818","https://openalex.org/W7202344424","https://openalex.org/W4404784060","https://openalex.org/W7223015661","https://openalex.org/W7170081662","https://openalex.org/W7128405121","https://openalex.org/W7128095460","https://openalex.org/W7221766575","https://openalex.org/W7127974521"],"abstract_inverted_index":{"Current":[0],"large":[1],"language":[2],"models":[3,75],"have":[4],"dangerous":[5],"capabilities,":[6],"which":[7],"are":[8],"likely":[9],"to":[10,23,68,76],"become":[11,77],"more":[12,78],"problematic":[13],"in":[14,115],"the":[15,34,48,105],"future.":[16],"Activation":[17],"steering":[18,38,63,69,91,99,110],"techniques":[19],"can":[20],"be":[21],"used":[22],"reduce":[24],"risks":[25],"from":[26],"these":[27],"capabilities.":[28],"In":[29,87],"this":[30],"paper,":[31],"we":[32,60,73],"investigate":[33],"efficacy":[35],"of":[36,50],"activation":[37],"for":[39,93],"broad":[40],"skills":[41,65],"and":[42,57,82],"multiple":[43,94],"behaviours.":[44,86],"First,":[45],"by":[46],"comparing":[47],"effects":[49],"reducing":[51],"performance":[52],"on":[53],"general":[54],"coding":[55],"ability":[56],"Python-specific":[58],"ability,":[59],"find":[61],"that":[62],"broader":[64],"is":[66,101,119],"competitive":[67],"narrower":[70],"skills.":[71],"Second,":[72],"steer":[74],"or":[79],"less":[80],"myopic":[81],"wealth-seeking,":[83],"among":[84],"other":[85,106],"our":[88],"experiments,":[89],"combining":[90],"vectors":[92,111],"different":[95,113],"behaviours":[96],"into":[97],"one":[98],"vector":[100],"largely":[102],"unsuccessful.":[103],"On":[104],"hand,":[107],"injecting":[108],"individual":[109],"at":[112],"places":[114],"a":[116],"model":[117],"simultaneously":[118],"promising.":[120]},"counts_by_year":[],"updated_date":"2026-10-12T09:40:04.860316","created_date":"2024-03-13T00:00:00"}
