{"id":"https://openalex.org/W4383996906","doi":"https://doi.org/10.48550/arxiv.2307.03637","title":"Discovering Variable Binding Circuitry with Desiderata","display_name":"Discovering Variable Binding Circuitry with Desiderata","publication_year":2023,"publication_date":"2023-07-07","ids":{"openalex":"https://openalex.org/W4383996906","doi":"https://doi.org/10.48550/arxiv.2307.03637"},"language":"en","primary_location":{"id":"pmh:oai:arXiv.org:2307.03637","doi":null,"is_oa":true,"landing_page_url":"http://arxiv.org/abs/2307.03637","pdf_url":"https://arxiv.org/pdf/2307.03637","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","is_traditional":true,"indexed_in":["arxiv","datacite"],"open_access":{"is_oa":true,"oa_status":"green","oa_url":"https://arxiv.org/pdf/2307.03637","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5145865735","display_name":"Xander Davies","orcid":null,"observed_orcids":[]},"institutions":[{"id":"https://openalex.org/I136199984","display_name":"Harvard University","ror":"https://ror.org/03vek6s52","country_code":"US","type":"education","lineage":["https://openalex.org/I136199984"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Davies, Xander","raw_affiliation_strings":["Harvard University"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Harvard University","institution_ids":["https://openalex.org/I136199984"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5074918225","display_name":"Max Nadeau","orcid":null,"observed_orcids":[]},"institutions":[{"id":"https://openalex.org/I136199984","display_name":"Harvard University","ror":"https://ror.org/03vek6s52","country_code":"US","type":"education","lineage":["https://openalex.org/I136199984"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Nadeau, Max","raw_affiliation_strings":["Harvard University"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Harvard University","institution_ids":["https://openalex.org/I136199984"]}]},{"author_position":"middle","author":{"id":null,"display_name":"Prakash, Nikhil","orcid":null,"observed_orcids":[]},"institutions":[{"id":"https://openalex.org/I12912129","display_name":"Northeastern University","ror":"https://ror.org/04t5xt781","country_code":"US","type":"education","lineage":["https://openalex.org/I12912129"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Prakash, Nikhil","raw_affiliation_strings":["Northeastern University"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Northeastern University","institution_ids":["https://openalex.org/I12912129"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5081132390","display_name":"Tamar Rott Shaham","orcid":"https://orcid.org/0000-0002-1455-2266","observed_orcids":["https://orcid.org/0000-0002-1455-2266"]},"institutions":[{"id":"https://openalex.org/I4405258931","display_name":"MIT Computer Science and Artificial Intelligence Laboratory","ror":"https://ror.org/01e85r662","country_code":"US","type":"facility","lineage":["https://openalex.org/I4405258931","https://openalex.org/I63966007"]},{"id":"https://openalex.org/I63966007","display_name":"Massachusetts Institute of Technology","ror":"https://ror.org/042nb2s44","country_code":"US","type":"education","lineage":["https://openalex.org/I63966007"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Shaham, Tamar Rott","raw_affiliation_strings":["MIT CSAIL"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"MIT CSAIL","institution_ids":["https://openalex.org/I4405258931","https://openalex.org/I63966007"]}]},{"author_position":"last","author":{"id":"https://openalex.org/A5033305045","display_name":"David Bau","orcid":"https://orcid.org/0000-0003-1744-6765","observed_orcids":["https://orcid.org/0000-0003-1744-6765"]},"institutions":[{"id":"https://openalex.org/I12912129","display_name":"Northeastern University","ror":"https://ror.org/04t5xt781","country_code":"US","type":"education","lineage":["https://openalex.org/I12912129"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Bau, David","raw_affiliation_strings":["Northeastern University"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Northeastern University","institution_ids":["https://openalex.org/I12912129"]}]}],"institutions":[],"countries_distinct_count":1,"institutions_distinct_count":4,"corresponding_author_ids":[],"corresponding_institution_ids":[],"apc_list":null,"apc_paid":null,"fwci":null,"has_fulltext":true,"cited_by_count":27,"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":0.9976000189781189,"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":0.9976000189781189,"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":0.0005000000237487257,"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"}},{"id":"https://openalex.org/T10743","display_name":"Software Testing and Debugging Techniques","score":0.0003000000142492354,"subfield":{"id":"https://openalex.org/subfields/1712","display_name":"Software"},"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/variable-binding","display_name":"variable binding","score":0.8939999938011169},{"id":"https://openalex.org/keywords/causal-mediation-analysis","display_name":"causal mediation analysis","score":0.7289999723434448},{"id":"https://openalex.org/keywords/desiderata","display_name":"desiderata","score":0.6970000267028809},{"id":"https://openalex.org/keywords/language-model-interpretability","display_name":"language model interpretability","score":0.6209999918937683},{"id":"https://openalex.org/keywords/arithmetic-problems","display_name":"arithmetic exercises","score":0.6079999804496765},{"id":"https://openalex.org/keywords/attention-heads","display_name":"attention heads","score":0.8230000138282776},{"id":"https://openalex.org/keywords/residual-stream","display_name":"residual stream","score":0.8830000162124634}],"concepts":[{"id":"https://openalex.org/C182365436","wikidata":"https://www.wikidata.org/wiki/Q50701","display_name":"Variable (mathematics)","level":2,"score":0.7380731105804443},{"id":"https://openalex.org/C48145219","wikidata":"https://www.wikidata.org/wiki/Q1335365","display_name":"Security token","level":2,"score":0.718429446220398},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.7071071267127991},{"id":"https://openalex.org/C177264268","wikidata":"https://www.wikidata.org/wiki/Q1514741","display_name":"Set (abstract data type)","level":2,"score":0.6811645030975342},{"id":"https://openalex.org/C45374587","wikidata":"https://www.wikidata.org/wiki/Q12525525","display_name":"Computation","level":2,"score":0.534363865852356},{"id":"https://openalex.org/C179420905","wikidata":"https://www.wikidata.org/wiki/Q223871","display_name":"Mediation","level":2,"score":0.46602579951286316},{"id":"https://openalex.org/C80444323","wikidata":"https://www.wikidata.org/wiki/Q2878974","display_name":"Theoretical computer science","level":1,"score":0.46438464522361755},{"id":"https://openalex.org/C11671645","wikidata":"https://www.wikidata.org/wiki/Q5054567","display_name":"Causal model","level":2,"score":0.4470328688621521},{"id":"https://openalex.org/C199360897","wikidata":"https://www.wikidata.org/wiki/Q9143","display_name":"Programming language","level":1,"score":0.42788994312286377},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.3861120045185089},{"id":"https://openalex.org/C204321447","wikidata":"https://www.wikidata.org/wiki/Q30642","display_name":"Natural language processing","level":1,"score":0.339846670627594},{"id":"https://openalex.org/C11413529","wikidata":"https://www.wikidata.org/wiki/Q8366","display_name":"Algorithm","level":1,"score":0.32905077934265137},{"id":"https://openalex.org/C94375191","wikidata":"https://www.wikidata.org/wiki/Q11205","display_name":"Arithmetic","level":1,"score":0.3230815529823303},{"id":"https://openalex.org/C33923547","wikidata":"https://www.wikidata.org/wiki/Q395","display_name":"Mathematics","level":0,"score":0.15797686576843262},{"id":"https://openalex.org/C17744445","wikidata":"https://www.wikidata.org/wiki/Q36442","display_name":"Political science","level":0,"score":0.0},{"id":"https://openalex.org/C134306372","wikidata":"https://www.wikidata.org/wiki/Q7754","display_name":"Mathematical analysis","level":1,"score":0.0},{"id":"https://openalex.org/C105795698","wikidata":"https://www.wikidata.org/wiki/Q12483","display_name":"Statistics","level":1,"score":0.0},{"id":"https://openalex.org/C199539241","wikidata":"https://www.wikidata.org/wiki/Q7748","display_name":"Law","level":1,"score":0.0},{"id":"https://openalex.org/C38652104","wikidata":"https://www.wikidata.org/wiki/Q3510521","display_name":"Computer security","level":1,"score":0.0}],"mesh":[],"locations_count":2,"locations":[{"id":"pmh:oai:arXiv.org:2307.03637","doi":null,"is_oa":true,"landing_page_url":"http://arxiv.org/abs/2307.03637","pdf_url":"https://arxiv.org/pdf/2307.03637","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.2307.03637","doi":"https://doi.org/10.48550/arxiv.2307.03637","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2307.03637","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:2307.03637","doi":null,"is_oa":true,"landing_page_url":"http://arxiv.org/abs/2307.03637","pdf_url":"https://arxiv.org/pdf/2307.03637","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/F4320315389","display_name":"Open Philanthropy Project","ror":"https://ror.org/004d1k391"}],"has_content":{"grobid_xml":true,"pdf":true},"content_urls":{"pdf":"https://content.openalex.org/works/W4383996906.pdf","grobid_xml":"https://content.openalex.org/works/W4383996906.grobid-xml"},"referenced_works_count":6,"referenced_works":["https://openalex.org/W2771655537","https://openalex.org/W3010694149","https://openalex.org/W4249352616","https://openalex.org/W4307413986","https://openalex.org/W4323557327","https://openalex.org/W4378976798"],"related_works":["https://openalex.org/W4416037884","https://openalex.org/W7160952012","https://openalex.org/W4412886783","https://openalex.org/W4404782280","https://openalex.org/W7202150508","https://openalex.org/W7214106579","https://openalex.org/W7164919854","https://openalex.org/W7220661578","https://openalex.org/W7166858631","https://openalex.org/W4417091950"],"abstract_inverted_index":{"Recent":[0],"work":[1],"has":[2],"shown":[3],"that":[4,62],"computation":[5],"in":[6,80,107],"language":[7],"models":[8],"may":[9],"be":[10],"human-understandable,":[11],"with":[12],"successful":[13],"efforts":[14],"to":[15,36,73,96],"localize":[16],"and":[17,23,104],"intervene":[18],"on":[19],"both":[20],"single-unit":[21],"features":[22],"input-output":[24],"circuits.":[25],"Here,":[26],"we":[27,69],"introduce":[28],"an":[29],"approach":[30],"which":[31,82],"extends":[32],"causal":[33,55],"mediation":[34],"experiments":[35],"automatically":[37,74],"identify":[38],"model":[39,59],"components":[40,60],"responsible":[41],"for":[42,86],"performing":[43],"a":[44,50,65],"specific":[45],"subtask":[46],"by":[47],"solely":[48],"specifying":[49],"set":[51],"of":[52,57,67],"\\textit{desiderata},":[53],"or":[54],"attributes":[56],"the":[58,102,108],"executing":[61],"subtask.":[63],"As":[64],"proof":[66],"concept,":[68],"apply":[70],"our":[71],"method":[72,91],"discover":[75],"shared":[76],"\\textit{variable":[77],"binding":[78,95],"circuitry}":[79],"LLaMA-13B,":[81],"retrieves":[83],"variable":[84,94],"values":[85],"multiple":[87],"arithmetic":[88],"tasks.":[89],"Our":[90],"successfully":[92],"localizes":[93],"only":[97],"9":[98],"attention":[99],"heads":[100],"(of":[101],"1.6k)":[103],"one":[105],"MLP":[106],"final":[109],"token's":[110],"residual":[111],"stream.":[112]},"counts_by_year":[{"year":2026,"cited_by_count":13},{"year":2025,"cited_by_count":7},{"year":2024,"cited_by_count":3},{"year":2023,"cited_by_count":4}],"updated_date":"2026-11-07T11:25:07.637180","created_date":"2023-07-12T00:00:00"}