{"id":"https://openalex.org/W7202200355","doi":"https://doi.org/10.21203/rs.3.rs-10666101/v1","title":"When Does a Readout Reflect Computation? Dose-Response Interventions in Continuous Thought Models","display_name":"When Does a Readout Reflect Computation? Dose-Response Interventions in Continuous Thought Models","publication_year":2026,"publication_date":"2026-08-12","ids":{"openalex":"https://openalex.org/W7202200355","doi":"https://doi.org/10.21203/rs.3.rs-10666101/v1"},"language":"en","primary_location":{"id":"doi:10.21203/rs.3.rs-10666101/v1","doi":"https://doi.org/10.21203/rs.3.rs-10666101/v1","is_oa":true,"landing_page_url":"https://doi.org/10.21203/rs.3.rs-10666101/v1","pdf_url":"https://www.researchsquare.com/article/rs-10666101/latest.pdf","source":{"id":"https://openalex.org/S4306525896","display_name":"Research Square","issn_l":null,"issn":null,"is_oa":false,"is_in_doaj":false,"is_core":false,"listed_in":[],"host_organization":null,"host_organization_name":null,"host_organization_lineage":[],"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":"Research Square","raw_type":"posted-content"},"parent_works":[],"lineage":["https://openalex.org/W7202200355"],"child_works_count":0,"type":"preprint","is_traditional":true,"indexed_in":["crossref"],"open_access":{"is_oa":true,"oa_status":"green","oa_url":"https://www.researchsquare.com/article/rs-10666101/latest.pdf","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5133115701","display_name":"ChaeWoo Son","orcid":"https://orcid.org/0009-0009-6108-2373","observed_orcids":["https://orcid.org/0009-0009-6108-2373"]},"institutions":[],"countries":[],"is_corresponding":true,"raw_author_name":"ChaeWoo Son","raw_affiliation_strings":["Independent Researcher"],"raw_orcid":"https://orcid.org/0009-0009-6108-2373","affiliations":[{"raw_affiliation_string":"Independent Researcher","institution_ids":[]}]}],"institutions":[],"countries_distinct_count":0,"institutions_distinct_count":0,"corresponding_author_ids":["https://openalex.org/A5133115701"],"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":0.9955999851226807,"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":0.9955999851226807,"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":0.0020000000949949026,"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":0.0005000000237487257,"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/causal-intervention","display_name":"causal intervention","score":0.3880000114440918},{"id":"https://openalex.org/keywords/latent-space","display_name":"latent space","score":0.5609999895095825}],"concepts":[{"id":"https://openalex.org/C45374587","wikidata":"https://www.wikidata.org/wiki/Q12525525","display_name":"Computation","level":2,"score":0.6211000084877014},{"id":"https://openalex.org/C2776214188","wikidata":"https://www.wikidata.org/wiki/Q408386","display_name":"Inference","level":2,"score":0.5012999773025513},{"id":"https://openalex.org/C2777601683","wikidata":"https://www.wikidata.org/wiki/Q6499736","display_name":"Vocabulary","level":2,"score":0.4717000126838684},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.46160000562667847},{"id":"https://openalex.org/C31395832","wikidata":"https://www.wikidata.org/wiki/Q1318674","display_name":"Testbed","level":2,"score":0.4000999927520752},{"id":"https://openalex.org/C168167062","wikidata":"https://www.wikidata.org/wiki/Q1117970","display_name":"Component (thermodynamics)","level":2,"score":0.37880000472068787},{"id":"https://openalex.org/C94915269","wikidata":"https://www.wikidata.org/wiki/Q1834857","display_name":"Detector","level":2,"score":0.3531999886035919},{"id":"https://openalex.org/C11413529","wikidata":"https://www.wikidata.org/wiki/Q8366","display_name":"Algorithm","level":1,"score":0.3434999883174896},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.34049999713897705},{"id":"https://openalex.org/C191795146","wikidata":"https://www.wikidata.org/wiki/Q3878446","display_name":"Norm (philosophy)","level":2,"score":0.3264000117778778},{"id":"https://openalex.org/C2780009758","wikidata":"https://www.wikidata.org/wiki/Q6804172","display_name":"Measure (data warehouse)","level":2,"score":0.28209999203681946},{"id":"https://openalex.org/C80444323","wikidata":"https://www.wikidata.org/wiki/Q2878974","display_name":"Theoretical computer science","level":1,"score":0.25920000672340393},{"id":"https://openalex.org/C25343380","wikidata":"https://www.wikidata.org/wiki/Q277521","display_name":"Relation (database)","level":2,"score":0.2547999918460846},{"id":"https://openalex.org/C158600405","wikidata":"https://www.wikidata.org/wiki/Q5054566","display_name":"Causal inference","level":2,"score":0.25429999828338623}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.21203/rs.3.rs-10666101/v1","doi":"https://doi.org/10.21203/rs.3.rs-10666101/v1","is_oa":true,"landing_page_url":"https://doi.org/10.21203/rs.3.rs-10666101/v1","pdf_url":"https://www.researchsquare.com/article/rs-10666101/latest.pdf","source":{"id":"https://openalex.org/S4306525896","display_name":"Research Square","issn_l":null,"issn":null,"is_oa":false,"is_in_doaj":false,"is_core":false,"listed_in":[],"host_organization":null,"host_organization_name":null,"host_organization_lineage":[],"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":"Research Square","raw_type":"posted-content"}],"best_oa_location":{"id":"doi:10.21203/rs.3.rs-10666101/v1","doi":"https://doi.org/10.21203/rs.3.rs-10666101/v1","is_oa":true,"landing_page_url":"https://doi.org/10.21203/rs.3.rs-10666101/v1","pdf_url":"https://www.researchsquare.com/article/rs-10666101/latest.pdf","source":{"id":"https://openalex.org/S4306525896","display_name":"Research Square","issn_l":null,"issn":null,"is_oa":false,"is_in_doaj":false,"is_core":false,"listed_in":[],"host_organization":null,"host_organization_name":null,"host_organization_lineage":[],"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":"Research Square","raw_type":"posted-content"},"sustainable_development_goals":[],"study_designs":[],"awards":[],"funders":[],"has_content":{"pdf":true,"grobid_xml":true},"content_urls":{"pdf":"https://content.openalex.org/works/W7202200355.pdf","grobid_xml":"https://content.openalex.org/works/W7202200355.grobid-xml"},"referenced_works_count":4,"referenced_works":["https://openalex.org/W4327526719","https://openalex.org/W4405254469","https://openalex.org/W4416037467","https://openalex.org/W7151605109"],"related_works":["https://openalex.org/W7128478628","https://openalex.org/W7128064915","https://openalex.org/W7164548276","https://openalex.org/W7202355060","https://openalex.org/W7202166788","https://openalex.org/W7165106955","https://openalex.org/W7222489584","https://openalex.org/W7168251903","https://openalex.org/W7172483257","https://openalex.org/W2407489385"],"abstract_inverted_index":{"Abstract":[0],"Chain":[1],"of":[2,96,104,175,220,268,289,298,304],"Continuous":[3],"Thought":[4],"(Coconut)":[5],"performs":[6],"computation":[7,47,160],"in":[8,39,74,81,216,242,291],"a":[9,37,67,79,89,133,153,206,286,295,301,318],"continuous":[10],"latent":[11,30,75],"space,":[12,34],"which":[13,82,209],"makes":[14],"its":[15],"intermediate":[16],"steps":[17],"difficult":[18],"to":[19,24,77,208],"monitor":[20],"directly.":[21],"Work":[22],"seeking":[23],"interpret":[25],"it":[26,127,269],"relies":[27],"on":[28,137,190,201,228,270],"projecting":[29],"states":[31,76],"into":[32,118],"vocabulary":[33],"and":[35,124,130,140,152,158,193,230,239,258],"treats":[36],"change":[38,238],"that":[40,44,53,92,249,275],"readout":[41,65,84,110,157,169,211,234,256,302],"as":[42,66,312],"evidence":[43],"the":[45,63,83,94,114,119,121,125,143,156,159,165,178,184,198,202,210,243,255,262,271],"underlying":[46],"has":[48,54],"changed":[49],"\u2014":[50,265],"an":[51],"inference":[52],"not":[55,181,213],"itself":[56],"been":[57],"characterized.":[58],"We":[59,112],"measure":[60],"it.":[61],"Using":[62],"J-lens":[64],"training":[68],"signal,":[69],"we":[70],"plant":[71],"bridge":[72],"entities":[73],"obtain":[78],"testbed":[80],"is":[85,187,194,277],"controlled,":[86],"then":[87],"apply":[88],"dose-response":[90,313],"protocol":[91],"fixes":[93],"magnitude":[95,229,297],"each":[97],"intervention.":[98],"Bridges":[99],"are":[100,222],"planted":[101],"without":[102],"loss":[103],"accuracy":[105],"(100%":[106],"answer":[107,179,185,237],"accuracy,":[108],"0.999":[109],"probability).":[111],"decompose":[113],"causal":[115,263],"intervention":[116,252],"delta":[117],"component":[120,126],"lens":[122],"detects":[123],"does":[128,180,212],"not,":[129],"control":[131],"with":[132,294],"random":[134,225],"direction":[135,207,226],"matched":[136,227],"both":[138],"norm":[139],"lens-invisibility.":[141],"Across":[142],"two":[144],"relation":[145,192,273],"types":[146],"used":[147],"for":[148],"intervention,":[149],"three":[150],"seeds":[151],"2B":[154],"model,":[155],"have":[161],"different":[162],"thresholds":[163],"along":[164,205],"same":[166,244],"direction.":[167],"The":[168,224],"flips":[170],"at":[171,254,266,279,317],"relative":[172],"perturbation":[173],"magnitudes":[174],"0.13-0.16,":[176],"but":[177],"move":[182],"there;":[183],"threshold":[186,257,264,276,303],"2.03x":[188],"higher":[189],"one":[191,272],"never":[195],"reached":[196,278],"before":[197],"model":[199],"breaks":[200],"other.":[203],"Conversely,":[204],"respond":[214],"once":[215],"1,800":[217,245],"trials,":[218],"96-97%":[219],"answers":[221],"transplanted.":[223],"invisibility":[231],"produced":[232,285],"no":[233,236,240],"change,":[235],"breakage":[241],"trials.":[246],"This":[247],"implies":[248],"minimum-norm":[250],"readout-targeted":[251],"stops":[253],"therefore":[259],"measures":[260],"below":[261],"half":[267],"where":[274],"all.":[280],"Our":[281],"own":[282],"initial":[283],"experiment":[284],"false":[287],"null":[288],"0/90":[290],"this":[292],"trap,":[293],"stopping":[296],"0.119":[299],"against":[300],"0.125.":[305],"Readout-based":[306],"interpretability":[307],"claims":[308],"should":[309],"be":[310],"reported":[311],"curves":[314],"rather":[315],"than":[316],"single":[319],"magnitude.":[320]},"counts_by_year":[],"updated_date":"2026-11-09T13:22:03.289127","created_date":"2026-08-12T00:00:00"}