{"id":"https://openalex.org/W7225189593","doi":"https://doi.org/10.48550/arxiv.2611.02503","title":"Towards Multi-Value Alignment via Value Localization in LLMs","display_name":"Towards Multi-Value Alignment via Value Localization in LLMs","publication_year":2026,"publication_date":"2026-11-02","ids":{"openalex":"https://openalex.org/W7225189593","doi":"https://doi.org/10.48550/arxiv.2611.02503"},"language":"en","primary_location":{"id":"pmh:oai:arXiv.org:2611.02503","doi":null,"is_oa":true,"landing_page_url":"https://arxiv.org/abs/2611.02503","pdf_url":"https://arxiv.org/pdf/2611.02503","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/2611.02503","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5160491933","display_name":"Xueqi Ma","orcid":null,"observed_orcids":[]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Ma, Xueqi","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5111270283","display_name":"Yanbei Jiang","orcid":"https://orcid.org/0009-0006-6987-4138","observed_orcids":["https://orcid.org/0009-0006-6987-4138"]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Jiang, Yanbei","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5160531776","display_name":"Xingjun Ma","orcid":null,"observed_orcids":[]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Ma, Xingjun","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5101609697","display_name":"James Bailey","orcid":"https://orcid.org/0000-0002-3769-3811","observed_orcids":["https://orcid.org/0000-0002-3769-3811"]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Bailey, James","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5070030398","display_name":"Sarah Erfani","orcid":"https://orcid.org/0000-0003-0885-0643","observed_orcids":["https://orcid.org/0000-0003-0885-0643"]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Erfani, Sarah M.","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/T12026","display_name":"Explainable Artificial Intelligence (XAI)","score":0.9624000191688538,"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":0.9624000191688538,"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":0.020600000396370888,"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":0.0052999998442828655,"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/attention-heads","display_name":"attention heads","score":0.6970000267028809},{"id":"https://openalex.org/keywords/mechanistic-interpretability","display_name":"mechanistic interpretability","score":0.9240000247955322},{"id":"https://openalex.org/keywords/probing","display_name":"probing","score":0.734000027179718},{"id":"https://openalex.org/keywords/activation-steering","display_name":"activation steering","score":0.9070000052452087},{"id":"https://openalex.org/keywords/value-conflicts","display_name":"value conflicts","score":0.5059999823570251},{"id":"https://openalex.org/keywords/harmlessness","display_name":"harmlessness","score":0.6690000295639038},{"id":"https://openalex.org/keywords/honesty","display_name":"honesty","score":0.9739999771118164},{"id":"https://openalex.org/keywords/stylistic-preferences","display_name":"stylistic preferences","score":0.800000011920929},{"id":"https://openalex.org/keywords/large-language-models","display_name":"large language models","score":0.6850000023841858}],"concepts":[{"id":"https://openalex.org/C2781067378","wikidata":"https://www.wikidata.org/wiki/Q17027399","display_name":"Interpretability","level":2,"score":0.9602000117301941},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.6033999919891357},{"id":"https://openalex.org/C12713177","wikidata":"https://www.wikidata.org/wiki/Q1900281","display_name":"Perspective (graphical)","level":2,"score":0.5856999754905701},{"id":"https://openalex.org/C2776291640","wikidata":"https://www.wikidata.org/wiki/Q2912517","display_name":"Value (mathematics)","level":2,"score":0.5430999994277954},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.5139999985694885},{"id":"https://openalex.org/C119857082","wikidata":"https://www.wikidata.org/wiki/Q2539","display_name":"Machine learning","level":1,"score":0.4036000072956085},{"id":"https://openalex.org/C2775924081","wikidata":"https://www.wikidata.org/wiki/Q55608371","display_name":"Control (management)","level":2,"score":0.3499000072479248},{"id":"https://openalex.org/C2776760102","wikidata":"https://www.wikidata.org/wiki/Q5139990","display_name":"Code (set theory)","level":3,"score":0.335099995136261},{"id":"https://openalex.org/C89611455","wikidata":"https://www.wikidata.org/wiki/Q6804646","display_name":"Mechanism (biology)","level":2,"score":0.3172000050544739},{"id":"https://openalex.org/C2780791683","wikidata":"https://www.wikidata.org/wiki/Q846785","display_name":"Action (physics)","level":2,"score":0.28189998865127563},{"id":"https://openalex.org/C11671645","wikidata":"https://www.wikidata.org/wiki/Q5054567","display_name":"Causal model","level":2,"score":0.27889999747276306}],"mesh":[],"locations_count":2,"locations":[{"id":"pmh:oai:arXiv.org:2611.02503","doi":null,"is_oa":true,"landing_page_url":"https://arxiv.org/abs/2611.02503","pdf_url":"https://arxiv.org/pdf/2611.02503","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.2611.02503","doi":"https://doi.org/10.48550/arxiv.2611.02503","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2611.02503","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:2611.02503","doi":null,"is_oa":true,"landing_page_url":"https://arxiv.org/abs/2611.02503","pdf_url":"https://arxiv.org/pdf/2611.02503","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":false,"grobid_xml":false},"content_urls":null,"referenced_works_count":0,"referenced_works":[],"related_works":["https://openalex.org/W7106698645","https://openalex.org/W7127598118","https://openalex.org/W7139941223","https://openalex.org/W4406031821","https://openalex.org/W6948375364","https://openalex.org/W4415237778","https://openalex.org/W7213437861","https://openalex.org/W7128489082","https://openalex.org/W7166839968","https://openalex.org/W7214597055"],"abstract_inverted_index":{"Aligning":[0],"large":[1],"language":[2],"models":[3],"(LLMs)":[4],"with":[5,68,134,174],"multiple,":[6],"potentially":[7],"conflicting":[8],"human":[9,73],"values":[10,74,136,140],"remains":[11],"challenging,":[12],"as":[13],"existing":[14],"multi-objective":[15],"alignment":[16,39,133,155],"methods":[17],"often":[18],"require":[19],"costly":[20],"fine-tuning":[21],"and":[22,78,92,160,184],"provide":[23],"limited":[24],"insight":[25],"into":[26],"the":[27,143],"internal":[28],"mechanisms":[29],"underlying":[30,144],"value-related":[31],"behaviors.":[32],"In":[33],"this":[34],"work,":[35],"we":[36,60,123],"investigate":[37],"multi-value":[38,161,176],"from":[40],"a":[41,55,179],"mechanistic":[42,172],"interpretability":[43,173],"perspective":[44],"by":[45,148],"identifying":[46],"value-sensitive":[47],"attention":[48,65],"heads":[49,66],"(value":[50],"heads)":[51],"in":[52],"LLMs.":[53],"Using":[54],"probing-based":[56],"value":[57,96,154],"localization":[58],"framework,":[59],"discover":[61],"sparse":[62],"sets":[63],"of":[64,115,139],"associated":[67],"diverse":[69],"dimensions,":[70],"spanning":[71],"both":[72],"(e.g.,":[75,81],"harmlessness,":[76],"honesty)":[77],"stylistic":[79],"preferences":[80],"coherence,":[82],"complexity).":[83],"Analyses":[84],"across":[85,152],"multiple":[86,153],"LLM":[87,186],"families":[88],"reveal":[89],"functional":[90,117],"specialization":[91],"structured":[93],"interactions":[94],"among":[95],"heads,":[97],"including":[98],"conflicts":[99],"between":[100],"certain":[101],"values.":[102],"Targeted":[103],"head":[104],"ablations":[105],"further":[106],"induce":[107],"substantial,":[108],"value-specific":[109],"behavioral":[110],"changes,":[111],"providing":[112],"causal":[113],"evidence":[114],"their":[116],"importance.":[118],"Building":[119],"on":[120],"these":[121],"findings,":[122],"propose":[124],"an":[125],"inference-time":[126],"activation":[127],"steering":[128],"strategy":[129],"that":[130],"enables":[131],"controllable":[132],"individual":[135],"or":[137],"combinations":[138],"without":[141],"updating":[142],"model":[145,167],"parameters,":[146],"supported":[147],"theoretical":[149],"analysis.":[150],"Experiments":[151],"benchmarks":[156],"demonstrate":[157],"effective":[158],"single-":[159],"control":[162],"while":[163],"largely":[164],"preserving":[165],"non-target":[166],"performance.":[168],"Our":[169],"findings":[170],"connect":[171],"flexible":[175],"alignment,":[177],"offering":[178],"foundation":[180],"for":[181],"more":[182],"interpretable":[183],"pluralistic":[185],"behavior.":[187],"The":[188],"code":[189],"is":[190],"available":[191],"at":[192],"https://github.com/sihuo-design/ValueHeads.":[193]},"counts_by_year":[],"updated_date":"2026-11-05T12:25:35.769452","created_date":"2026-11-04T00:00:00"}