{"id":"https://openalex.org/W7225575750","doi":"https://doi.org/10.5281/zenodo.24053283","title":"Neurons of a known censorship as an internal probe: a verified cascade audit of Qwen3.8-27B, complementary to an outside sensitivity prior","display_name":"Neurons of a known censorship as an internal probe: a verified cascade audit of Qwen3.8-27B, complementary to an outside sensitivity prior","publication_year":2026,"publication_date":"2026-11-08","ids":{"openalex":"https://openalex.org/W7225575750","doi":"https://doi.org/10.5281/zenodo.24053283"},"language":"en","primary_location":{"id":"doi:10.5281/zenodo.24053283","doi":"https://doi.org/10.5281/zenodo.24053283","is_oa":true,"landing_page_url":"https://doi.org/10.5281/zenodo.24053283","pdf_url":null,"source":{"id":"https://openalex.org/S4306400562","display_name":"Zenodo (CERN European Organization for Nuclear Research)","issn_l":null,"issn":null,"is_oa":true,"is_in_doaj":false,"is_core":false,"listed_in":[],"host_organization":"https://openalex.org/I67311998","host_organization_name":"European Organization for Nuclear Research","host_organization_lineage":["https://openalex.org/I67311998"],"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"},"volume":null,"type":"preprint","is_traditional":true,"indexed_in":["datacite"],"open_access":{"is_oa":true,"oa_status":"green","oa_url":"https://doi.org/10.5281/zenodo.24053283","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5063912357","display_name":"Valerio Dolci","orcid":null,"observed_orcids":[]},"institutions":[],"countries":[],"is_corresponding":true,"raw_author_name":"Dolci, Valerio","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]}],"institutions":[],"countries_distinct_count":0,"institutions_distinct_count":0,"corresponding_author_ids":["https://openalex.org/A5063912357"],"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":0.322299987077713,"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":0.322299987077713,"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":0.2759000062942505,"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":0.12809999287128448,"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/censorship","display_name":"censorship","score":0.6399999856948853},{"id":"https://openalex.org/keywords/taiwan-question","display_name":"Taiwan question","score":0.39899998903274536},{"id":"https://openalex.org/keywords/chinese-language-models","display_name":"Chinese language models","score":0.37299999594688416},{"id":"https://openalex.org/keywords/western-sahara","display_name":"Western Sahara","score":0.3959999978542328},{"id":"https://openalex.org/keywords/crimea","display_name":"Crimea","score":0.9559999704360962},{"id":"https://openalex.org/keywords/open-source-toolkit","display_name":"open-source software toolkit","score":0.335999995470047}],"concepts":[{"id":"https://openalex.org/C774472","wikidata":"https://www.wikidata.org/wiki/Q6760393","display_name":"Margin (machine learning)","level":2,"score":0.6593000292778015},{"id":"https://openalex.org/C137668524","wikidata":"https://www.wikidata.org/wiki/Q189813","display_name":"Censoring (clinical trials)","level":2,"score":0.5189999938011169},{"id":"https://openalex.org/C198531522","wikidata":"https://www.wikidata.org/wiki/Q485146","display_name":"Sample (material)","level":2,"score":0.5156000256538391},{"id":"https://openalex.org/C52930066","wikidata":"https://www.wikidata.org/wiki/Q543","display_name":"Censorship","level":2,"score":0.51419997215271},{"id":"https://openalex.org/C21200559","wikidata":"https://www.wikidata.org/wiki/Q7451068","display_name":"Sensitivity (control systems)","level":2,"score":0.45820000767707825},{"id":"https://openalex.org/C61423126","wikidata":"https://www.wikidata.org/wiki/Q187432","display_name":"Scripting language","level":2,"score":0.45750001072883606},{"id":"https://openalex.org/C28719098","wikidata":"https://www.wikidata.org/wiki/Q44946","display_name":"Point (geometry)","level":2,"score":0.4505999982357025},{"id":"https://openalex.org/C186282968","wikidata":"https://www.wikidata.org/wiki/Q1434261","display_name":"McNemar's test","level":2,"score":0.4487999975681305},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.42809998989105225},{"id":"https://openalex.org/C15744967","wikidata":"https://www.wikidata.org/wiki/Q9418","display_name":"Psychology","level":0,"score":0.3531999886035919},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.3476000130176544},{"id":"https://openalex.org/C199521495","wikidata":"https://www.wikidata.org/wiki/Q181487","display_name":"Audit","level":2,"score":0.33480000495910645},{"id":"https://openalex.org/C64869954","wikidata":"https://www.wikidata.org/wiki/Q1859747","display_name":"False positive paradox","level":2,"score":0.3314000070095062},{"id":"https://openalex.org/C105795698","wikidata":"https://www.wikidata.org/wiki/Q12483","display_name":"Statistics","level":1,"score":0.3237999975681305},{"id":"https://openalex.org/C149782125","wikidata":"https://www.wikidata.org/wiki/Q160039","display_name":"Econometrics","level":1,"score":0.27649998664855957},{"id":"https://openalex.org/C44291984","wikidata":"https://www.wikidata.org/wiki/Q1074173","display_name":"Question answering","level":2,"score":0.26919999718666077},{"id":"https://openalex.org/C3019144022","wikidata":"https://www.wikidata.org/wiki/Q4124998","display_name":"Questions and answers","level":2,"score":0.26820001006126404},{"id":"https://openalex.org/C2779662365","wikidata":"https://www.wikidata.org/wiki/Q5416694","display_name":"Event (particle physics)","level":2,"score":0.2655999958515167},{"id":"https://openalex.org/C48798503","wikidata":"https://www.wikidata.org/wiki/Q877546","display_name":"Reputation","level":2,"score":0.26330000162124634},{"id":"https://openalex.org/C190812933","wikidata":"https://www.wikidata.org/wiki/Q28923","display_name":"Chart","level":2,"score":0.2596000134944916},{"id":"https://openalex.org/C48372109","wikidata":"https://www.wikidata.org/wiki/Q3913","display_name":"Binary number","level":2,"score":0.25870001316070557},{"id":"https://openalex.org/C177769412","wikidata":"https://www.wikidata.org/wiki/Q278090","display_name":"Prior probability","level":3,"score":0.25519999861717224}],"mesh":[],"locations_count":2,"locations":[{"id":"doi:10.5281/zenodo.24053283","doi":"https://doi.org/10.5281/zenodo.24053283","is_oa":true,"landing_page_url":"https://doi.org/10.5281/zenodo.24053283","pdf_url":null,"source":{"id":"https://openalex.org/S4306400562","display_name":"Zenodo (CERN European Organization for Nuclear Research)","issn_l":null,"issn":null,"is_oa":true,"is_in_doaj":false,"is_core":false,"listed_in":[],"host_organization":"https://openalex.org/I67311998","host_organization_name":"European Organization for Nuclear Research","host_organization_lineage":["https://openalex.org/I67311998"],"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"},{"id":"doi:10.5281/zenodo.24053284","doi":"https://doi.org/10.5281/zenodo.24053284","is_oa":true,"landing_page_url":"https://doi.org/10.5281/zenodo.24053284","pdf_url":null,"source":{"id":"https://openalex.org/S4306400562","display_name":"Zenodo (CERN European Organization for Nuclear Research)","issn_l":null,"issn":null,"is_oa":true,"is_in_doaj":false,"is_core":false,"listed_in":[],"host_organization":"https://openalex.org/I67311998","host_organization_name":"European Organization for Nuclear Research","host_organization_lineage":["https://openalex.org/I67311998"],"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":"doi:10.5281/zenodo.24053283","doi":"https://doi.org/10.5281/zenodo.24053283","is_oa":true,"landing_page_url":"https://doi.org/10.5281/zenodo.24053283","pdf_url":null,"source":{"id":"https://openalex.org/S4306400562","display_name":"Zenodo (CERN European Organization for Nuclear Research)","issn_l":null,"issn":null,"is_oa":true,"is_in_doaj":false,"is_core":false,"listed_in":[],"host_organization":"https://openalex.org/I67311998","host_organization_name":"European Organization for Nuclear Research","host_organization_lineage":["https://openalex.org/I67311998"],"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"},"sustainable_development_goals":[{"id":"https://metadata.un.org/sdg/16","score":0.5473999977111816,"display_name":"Peace, justice, and strong institutions"}],"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/W7162015586","https://openalex.org/W7223237006","https://openalex.org/W4414581015","https://openalex.org/W4414606756","https://openalex.org/W4416872504","https://openalex.org/W4412373306","https://openalex.org/W7134042486","https://openalex.org/W4417301673","https://openalex.org/W3196681588","https://openalex.org/W7084502838"],"abstract_inverted_index":{"We":[0],"start":[1],"from":[2,33],"one":[3],"censored":[4,31,130,159,247,280],"behaviour":[5],"we":[6],"already":[7],"know":[8],"(Qwen3.8-27B":[9],"denies":[10],"verifiable":[11],"facts":[12],"about":[13],"Taiwan":[14,61],"in":[15,22,37,108,131,139,249,309,315],"short":[16],"yes/no":[17],"answers":[18,64,120,145,157,190,212,259],"while":[19,254],"admitting":[20],"them":[21,262],"free":[23],"text)":[24],"and":[25,92,100,229,236,295,311,317,326,345],"use":[26],"it":[27],"to":[28,89,233],"find":[29],"other":[30],"topics":[32,184,194,214],"inside":[34],"the":[35,56,60,115,127,147,174,181,189,200,215,222,225,243,264,268,273,278,302,340],"model,":[36],"a":[38,43,81,84,97,101,140,177,208,250,255,336],"cascade.":[39],"Activation\u00d7gradient":[40],"attribution":[41],"of":[42,51,55,104,133,143,154,163,188,261,277,320],"two-answer":[44],"margin":[45],"selects":[46],"32":[47,269,282],"MLP":[48],"neurons":[49,76,271,284],"(32":[50],"1.1":[52],"M;":[53],"0.0006%":[54],"weights)":[57],"that":[58],"carry":[59],"script":[62],"(held-out":[63],"11/19":[65],"\u2192":[66],"17/19":[67],"factual":[68],"when":[69,176],"ablated,":[70],"exact":[71],"McNemar":[72],"p=0.03;":[73],"64":[74],"random":[75,141,283],"do":[77],"nothing).":[78],"Used":[79],"as":[80,218,291,335],"probe":[82],"over":[83],"neutral":[85],"battery,":[86],"they":[87],"point":[88],"contested":[90],"territories":[91],"China-aligned":[93],"governments;":[94],"this":[95],"guides":[96],"larger":[98],"battery":[99],"broader":[102],"map":[103,128],"censorship":[105,274],"neurons,":[106],"which":[107],"turn":[109],"ranks":[110],"topic":[111],"lists":[112],"written":[113],"by":[114,126,242,330],"model":[116,179,258],"itself.":[117],"Among":[118],"high-scoring":[119],"not":[121,219],"judged":[122,146],"before,":[123],"those":[124],"flagged":[125,156],"are":[129,158,195,246],"29%":[132],"cases":[134,240],"(86/293)":[135],"against":[136,161,197],"3.4%":[137],"(13/388)":[138],"sample":[142],"keyword-free":[144],"same":[148,182],"way;":[149],"among":[150],"model-listed":[151],"topics,":[152],"42":[153],"260":[155],"(16%),":[160],"none":[162],"60":[164],"low-scoring":[165],"ones.":[166],"An":[167],"outside":[168],"prior":[169,216],"does":[170],"better":[171],"at":[172],"finding":[173],"obvious:":[175],"non-Chinese":[178,256],"rates":[180,217],"listed":[183],"for":[185,199,275,285],"sensitivity,":[186],"41%":[187],"on":[191,213,263,297],"its":[192],"top":[193],"censored,":[196],"16%":[198],"neuron":[201],"probe.":[202],"The":[203,287],"probe's":[204],"distinct":[205],"contribution":[206],"is":[207,289],"non-obvious":[209],"tail":[210],"(censored":[211],"sensitive,":[220],"e.g.":[221],"YJ-18":[223],"missile,":[224],"director":[226],"Diao":[227],"Yinan)":[228],"sovereignty":[230],"scripts":[231],"applied":[232],"Western":[234],"Sahara":[235],"Crimea.":[237],"Of":[238],"28":[239],"surfaced":[241],"probes,":[244],"24":[245],"again":[248],"fresh":[251],"bf16":[252],"verification,":[253],"reference":[257],"all":[260],"merits;":[265],"switching":[266],"off":[267],"seed":[270],"removes":[272],"19":[276],"29":[279],"candidates,":[281],"1.":[286],"procedure":[288],"packaged":[290],"an":[292],"open":[293],"toolkit":[294],"rerun":[296],"Qwen3-4B.":[298],"Pre-registered":[299],"checks":[300],"confirm":[301],"core":[303],"(harmful-request":[304],"refusals":[305],"198/200":[306,312],"vs":[307,313],"200/200":[308],"English":[310],"199/200":[314],"Chinese)":[316],"withdrew":[318],"two":[319],"our":[321],"earlier":[322],"claims.":[323],"Experiments,":[324],"code":[325],"drafts":[327],"were":[328],"produced":[329],"Claude":[331],"Opus":[332],"5.5":[333],"working":[334],"coding":[337],"agent":[338],"under":[339],"author's":[341],"direction.":[342],"Code,":[343],"data":[344],"pre-registrations:":[346],"https://github.com/ValerioDolci/frontier-weight-edit":[347],"(folder":[348],"experiments/qwen27b-censorship).":[349]},"counts_by_year":[],"updated_date":"2026-11-10T11:39:43.021295","created_date":"2026-11-09T00:00:00"}
