{"id":"https://openalex.org/W7220909013","doi":"https://doi.org/10.48550/arxiv.2610.08200","title":"Where Do LLMs Retrieve Network Information? Identifying the Responsible Heads and Neurons","display_name":"Where Do LLMs Retrieve Network Information? Identifying the Responsible Heads and Neurons","publication_year":2026,"publication_date":"2026-10-06","ids":{"openalex":"https://openalex.org/W7220909013","doi":"https://doi.org/10.48550/arxiv.2610.08200"},"language":"en","primary_location":{"id":"pmh:oai:arXiv.org:2610.08200","doi":null,"is_oa":true,"landing_page_url":"https://arxiv.org/abs/2610.08200","pdf_url":"https://arxiv.org/pdf/2610.08200","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/2610.08200","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5096780386","display_name":"Md Abdul Kadir","orcid":null,"observed_orcids":[]},"institutions":[{"id":"https://openalex.org/I129877168","display_name":"Carl von Ossietzky Universit\u00e4t Oldenburg","ror":"https://ror.org/033n9gh91","country_code":"DE","type":"education","lineage":["https://openalex.org/I129877168"]},{"id":"https://openalex.org/I33256026","display_name":"German Research Centre for Artificial Intelligence","ror":"https://ror.org/01ayc5b57","country_code":"DE","type":"funder","lineage":["https://openalex.org/I33256026"]}],"countries":["DE"],"is_corresponding":false,"raw_author_name":"Kadir, Md Abdul","raw_affiliation_strings":["German Research Center for Artificial Intelligence","University of Oldenburg, Oldenburg, Germany"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"German Research Center for Artificial Intelligence","institution_ids":["https://openalex.org/I33256026"]},{"raw_affiliation_string":"University of Oldenburg, Oldenburg, Germany","institution_ids":["https://openalex.org/I129877168"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5139990335","display_name":"Md Mohasin Hossain","orcid":null,"observed_orcids":[]},"institutions":[{"id":"https://openalex.org/I33256026","display_name":"German Research Centre for Artificial Intelligence","ror":"https://ror.org/01ayc5b57","country_code":"DE","type":"funder","lineage":["https://openalex.org/I33256026"]},{"id":"https://openalex.org/I91712215","display_name":"Saarland University","ror":"https://ror.org/01jdpyv68","country_code":"DE","type":"education","lineage":["https://openalex.org/I91712215"]}],"countries":["DE"],"is_corresponding":false,"raw_author_name":"Hossain, Md Mohasin","raw_affiliation_strings":["German Research Center for Artificial Intelligence","Saarland University, Saarbr\u00fccken, Germany"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"German Research Center for Artificial Intelligence","institution_ids":["https://openalex.org/I33256026"]},{"raw_affiliation_string":"Saarland University, Saarbr\u00fccken, Germany","institution_ids":["https://openalex.org/I91712215"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5006937331","display_name":"Omair Shahzad Bhatti","orcid":"https://orcid.org/0000-0001-7983-2384","observed_orcids":["https://orcid.org/0000-0001-7983-2384"]},"institutions":[{"id":"https://openalex.org/I33256026","display_name":"German Research Centre for Artificial Intelligence","ror":"https://ror.org/01ayc5b57","country_code":"DE","type":"funder","lineage":["https://openalex.org/I33256026"]}],"countries":["DE"],"is_corresponding":false,"raw_author_name":"Bhatti, Omair Shahzad","raw_affiliation_strings":["German Research Center for Artificial Intelligence"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"German Research Center for Artificial Intelligence","institution_ids":["https://openalex.org/I33256026"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5032619772","display_name":"Abdulrahman Mohamed Selim","orcid":"https://orcid.org/0000-0002-4984-6686","observed_orcids":["https://orcid.org/0000-0002-4984-6686"]},"institutions":[{"id":"https://openalex.org/I129877168","display_name":"Carl von Ossietzky Universit\u00e4t Oldenburg","ror":"https://ror.org/033n9gh91","country_code":"DE","type":"education","lineage":["https://openalex.org/I129877168"]},{"id":"https://openalex.org/I33256026","display_name":"German Research Centre for Artificial Intelligence","ror":"https://ror.org/01ayc5b57","country_code":"DE","type":"funder","lineage":["https://openalex.org/I33256026"]}],"countries":["DE"],"is_corresponding":false,"raw_author_name":"Selim, Abdulrahman Mohamed","raw_affiliation_strings":["German Research Center for Artificial Intelligence","University of Oldenburg, Oldenburg, Germany"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"German Research Center for Artificial Intelligence","institution_ids":["https://openalex.org/I33256026"]},{"raw_affiliation_string":"University of Oldenburg, Oldenburg, Germany","institution_ids":["https://openalex.org/I129877168"]}]},{"author_position":"last","author":{"id":"https://openalex.org/A5069432042","display_name":"Daniel Sonntag","orcid":"https://orcid.org/0000-0002-8857-8709","observed_orcids":["https://orcid.org/0000-0002-8857-8709"]},"institutions":[{"id":"https://openalex.org/I129877168","display_name":"Carl von Ossietzky Universit\u00e4t Oldenburg","ror":"https://ror.org/033n9gh91","country_code":"DE","type":"education","lineage":["https://openalex.org/I129877168"]},{"id":"https://openalex.org/I33256026","display_name":"German Research Centre for Artificial Intelligence","ror":"https://ror.org/01ayc5b57","country_code":"DE","type":"funder","lineage":["https://openalex.org/I33256026"]}],"countries":["DE"],"is_corresponding":false,"raw_author_name":"Sonntag, Daniel","raw_affiliation_strings":["German Research Center for Artificial Intelligence","University of Oldenburg, Oldenburg, Germany"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"German Research Center for Artificial Intelligence","institution_ids":["https://openalex.org/I33256026"]},{"raw_affiliation_string":"University of Oldenburg, Oldenburg, Germany","institution_ids":["https://openalex.org/I129877168"]}]}],"institutions":[],"countries_distinct_count":1,"institutions_distinct_count":3,"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":0.444599986076355,"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":0.444599986076355,"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/T12026","display_name":"Explainable Artificial Intelligence (XAI)","score":0.41990000009536743,"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":0.06620000302791595,"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/large-language-models","display_name":"large language models","score":0.9900000095367432},{"id":"https://openalex.org/keywords/network-information-retrieval","display_name":"network information retrieval","score":0.925000011920929},{"id":"https://openalex.org/keywords/attention-heads","display_name":"attention heads","score":0.6819999814033508},{"id":"https://openalex.org/keywords/causal-ablation","display_name":"causal ablation","score":0.9179999828338623},{"id":"https://openalex.org/keywords/web-information-processing","display_name":"web information processing","score":0.5379999876022339},{"id":"https://openalex.org/keywords/model-interpretability","display_name":"model interpretability","score":0.414000004529953}],"concepts":[{"id":"https://openalex.org/C90329073","wikidata":"https://www.wikidata.org/wiki/Q914232","display_name":"Ask price","level":2,"score":0.6958000063896179},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.6445000171661377},{"id":"https://openalex.org/C177264268","wikidata":"https://www.wikidata.org/wiki/Q1514741","display_name":"Set (abstract data type)","level":2,"score":0.6218000054359436},{"id":"https://openalex.org/C165064840","wikidata":"https://www.wikidata.org/wiki/Q1321061","display_name":"Matching (statistics)","level":2,"score":0.590399980545044},{"id":"https://openalex.org/C2779343474","wikidata":"https://www.wikidata.org/wiki/Q3109175","display_name":"Context (archaeology)","level":2,"score":0.5817000269889832},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.5134000182151794},{"id":"https://openalex.org/C204321447","wikidata":"https://www.wikidata.org/wiki/Q30642","display_name":"Natural language processing","level":1,"score":0.41190001368522644},{"id":"https://openalex.org/C58489278","wikidata":"https://www.wikidata.org/wiki/Q1172284","display_name":"Data set","level":2,"score":0.4108000099658966},{"id":"https://openalex.org/C2780312720","wikidata":"https://www.wikidata.org/wiki/Q5689100","display_name":"Head (geology)","level":2,"score":0.3944999873638153},{"id":"https://openalex.org/C2780451532","wikidata":"https://www.wikidata.org/wiki/Q759676","display_name":"Task (project management)","level":2,"score":0.33629998564720154},{"id":"https://openalex.org/C195324797","wikidata":"https://www.wikidata.org/wiki/Q33742","display_name":"Natural language","level":2,"score":0.30799999833106995},{"id":"https://openalex.org/C15744967","wikidata":"https://www.wikidata.org/wiki/Q9418","display_name":"Psychology","level":0,"score":0.3075000047683716},{"id":"https://openalex.org/C192209626","wikidata":"https://www.wikidata.org/wiki/Q190909","display_name":"Focus (optics)","level":2,"score":0.3012000024318695},{"id":"https://openalex.org/C119857082","wikidata":"https://www.wikidata.org/wiki/Q2539","display_name":"Machine learning","level":1,"score":0.29820001125335693},{"id":"https://openalex.org/C117220453","wikidata":"https://www.wikidata.org/wiki/Q5172842","display_name":"Correlation","level":2,"score":0.29679998755455017},{"id":"https://openalex.org/C23123220","wikidata":"https://www.wikidata.org/wiki/Q816826","display_name":"Information retrieval","level":1,"score":0.2858000099658966},{"id":"https://openalex.org/C94915269","wikidata":"https://www.wikidata.org/wiki/Q1834857","display_name":"Detector","level":2,"score":0.2797999978065491},{"id":"https://openalex.org/C137293760","wikidata":"https://www.wikidata.org/wiki/Q3621696","display_name":"Language model","level":2,"score":0.27970001101493835},{"id":"https://openalex.org/C104122410","wikidata":"https://www.wikidata.org/wiki/Q1416406","display_name":"Network model","level":2,"score":0.26739999651908875},{"id":"https://openalex.org/C2778029271","wikidata":"https://www.wikidata.org/wiki/Q5421931","display_name":"Extension (predicate logic)","level":2,"score":0.26510000228881836},{"id":"https://openalex.org/C148524875","wikidata":"https://www.wikidata.org/wiki/Q6975395","display_name":"F1 score","level":2,"score":0.2605000138282776},{"id":"https://openalex.org/C138268822","wikidata":"https://www.wikidata.org/wiki/Q1051925","display_name":"Resolution (logic)","level":2,"score":0.2506999969482422}],"mesh":[],"locations_count":2,"locations":[{"id":"pmh:oai:arXiv.org:2610.08200","doi":null,"is_oa":true,"landing_page_url":"https://arxiv.org/abs/2610.08200","pdf_url":"https://arxiv.org/pdf/2610.08200","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.2610.08200","doi":"https://doi.org/10.48550/arxiv.2610.08200","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2610.08200","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:2610.08200","doi":null,"is_oa":true,"landing_page_url":"https://arxiv.org/abs/2610.08200","pdf_url":"https://arxiv.org/pdf/2610.08200","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/W7220909013.pdf","grobid_xml":"https://content.openalex.org/works/W7220909013.grobid-xml"},"referenced_works_count":2,"referenced_works":["https://openalex.org/W4406031134","https://openalex.org/W4414939355"],"related_works":["https://openalex.org/W7197046484","https://openalex.org/W7127084415","https://openalex.org/W6959727613","https://openalex.org/W7204130747","https://openalex.org/W7216323833","https://openalex.org/W4414968658","https://openalex.org/W7214738910","https://openalex.org/W7166873788","https://openalex.org/W6988012220","https://openalex.org/W7129742391"],"abstract_inverted_index":{"Large":[0],"language":[1],"models":[2],"(LLMs)":[3],"used":[4],"for":[5],"Web":[6],"information":[7,108,230],"processing":[8],"rely":[9],"on":[10,116],"retrieving":[11],"relationships":[12],"between":[13],"network":[14,107,229],"entities,":[15],"such":[16],"as":[17],"domain":[18],"names":[19],"and":[20,80,91,94,102,156],"IP":[21],"addresses,":[22],"but":[23],"the":[24,144,147,160,172,183,213],"internal":[25],"components":[26],"that":[27],"causally":[28,99,128,174,223],"enable":[29],"this":[30],"retrieval":[31,45,109],"remain":[32],"unclear,":[33],"hindering":[34],"targeted":[35],"safeguards":[36],"against":[37,77],"unsafe":[38],"actions.":[39],"Therefore,":[40],"we":[41,67,112],"investigate":[42],"whether":[43],"their":[44,75],"capabilities":[46],"can":[47],"be":[48],"localized":[49],"to":[50,105],"specific":[51],"attention":[52],"heads":[53,69,87,101,130],"or":[54],"individual":[55,103],"neurons":[56,84,104],"within":[57,85],"these":[58,86,210],"LLMs.":[59],"Across":[60],"five":[61,207],"LLMs":[62,226],"from":[63,97,231],"three":[64,205],"architecture":[65],"families,":[66],"identify":[68,159],"through":[70],"causal":[71,89,155],"ablation":[72,90],"screening,":[73],"test":[74],"selectivity":[76],"matched":[78],"negative":[79],"context-free":[81],"controls,":[82],"rank":[83],"using":[88],"correlational":[92,157],"measures,":[93],"leverage":[95],"features":[96],"both":[98],"validated":[100,129],"build":[106],"detectors,":[110],"which":[111],"evaluate":[113],"without":[114],"retraining":[115],"an":[117,217],"independent":[118,189],"reverse-DNS":[119],"dataset.":[120],"In":[121],"every":[122],"model,":[123,154],"a":[124,137],"small":[125],"set":[126],"of":[127,133,206,215,220],"(1--9":[131],"out":[132],"128--1152":[134],"candidates)":[135],"builds":[136],"detector":[138],"with":[139],"99.5--100\\%":[140],"held-out":[141],"accuracy.":[142],"At":[143],"neuron":[145,176],"level,":[146],"results":[148],"are":[149],"model-specific,":[150],"e.g.,":[151],"in":[152,170,204],"one":[153],"rankings":[158],"same":[161],"neuron,":[162],"matching":[163],"full-head":[164],"performance":[165],"(AUC":[166],"=":[167,180],"1.000),":[168],"while":[169,198],"another,":[171],"best":[173],"ranked":[175],"achieves":[177],"only":[178],"AUC":[179,195],"0.665":[181],"despite":[182],"full":[184],"head":[185,192],"reaching":[186],"1.000.":[187],"On":[188],"CAIDA":[190],"data,":[191],"detectors":[193,200],"maintain":[194],"$\\geq":[196],"0.971$,":[197],"single-neuron":[199],"fall":[201],"below":[202],"chance":[203],"models.":[208],"Overall,":[209],"findings":[211],"highlight":[212],"importance":[214],"choosing":[216],"appropriate":[218],"level":[219],"granularity":[221],"when":[222],"localizing":[224],"how":[225],"retrieve":[227],"structured":[228],"context.":[232]},"counts_by_year":[],"updated_date":"2026-11-08T12:12:28.789561","created_date":"2026-10-09T00:00:00"}