{"id":"https://openalex.org/W4403752327","doi":"https://doi.org/10.48550/arxiv.2409.13714","title":"TracrBench: Generating Interpretability Testbeds with Large Language Models","display_name":"TracrBench: Generating Interpretability Testbeds with Large Language Models","publication_year":2024,"publication_date":"2024-09-07","ids":{"openalex":"https://openalex.org/W4403752327","doi":"https://doi.org/10.48550/arxiv.2409.13714"},"language":"en","primary_location":{"id":"pmh:oai:arXiv.org:2409.13714","doi":null,"is_oa":true,"landing_page_url":"http://arxiv.org/abs/2409.13714","pdf_url":"https://arxiv.org/pdf/2409.13714","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","indexed_in":["arxiv","datacite"],"open_access":{"is_oa":true,"oa_status":"green","oa_url":"https://arxiv.org/pdf/2409.13714","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5107376168","display_name":"Hannes Thurnherr","orcid":"https://orcid.org/0009-0001-4554-3899","observed_orcids":["https://orcid.org/0009-0001-4554-3899"]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Thurnherr, Hannes","raw_affiliation_strings":["Independent , Zurich , CH"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Independent , Zurich , CH","institution_ids":[]}]},{"author_position":"last","author":{"id":"https://openalex.org/A5064744991","display_name":"J\u00e9r\u00e9my Scheurer","orcid":"https://orcid.org/0000-0002-6859-6029","observed_orcids":["https://orcid.org/0000-0002-6859-6029"]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Scheurer, J\u00e9r\u00e9my","raw_affiliation_strings":["Apollo Research , Lon- don , UK"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Apollo Research , Lon- don , UK","institution_ids":[]}]}],"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":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":1.0,"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":1.0,"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":0.0,"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":0.0,"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/rasp","display_name":"RASP","score":0.49799999594688416},{"id":"https://openalex.org/keywords/interpretability-evaluation","display_name":"interpretability evaluation","score":0.5799999833106995},{"id":"https://openalex.org/keywords/gpt-4-turbo","display_name":"GPT-4 Turbo","score":0.7739999890327454},{"id":"https://openalex.org/keywords/code-generation","display_name":"code generation","score":0.41100001335144043},{"id":"https://openalex.org/keywords/mechanistic-interpretability","display_name":"mechanistic interpretability","score":0.6949999928474426}],"concepts":[{"id":"https://openalex.org/C2781067378","wikidata":"https://www.wikidata.org/wiki/Q17027399","display_name":"Interpretability","level":2,"score":0.9316564798355103},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.7031100988388062},{"id":"https://openalex.org/C199360897","wikidata":"https://www.wikidata.org/wiki/Q9143","display_name":"Programming language","level":1,"score":0.41347670555114746},{"id":"https://openalex.org/C204321447","wikidata":"https://www.wikidata.org/wiki/Q30642","display_name":"Natural language processing","level":1,"score":0.40415075421333313},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.35787123441696167}],"mesh":[],"locations_count":2,"locations":[{"id":"pmh:oai:arXiv.org:2409.13714","doi":null,"is_oa":true,"landing_page_url":"http://arxiv.org/abs/2409.13714","pdf_url":"https://arxiv.org/pdf/2409.13714","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.2409.13714","doi":"https://doi.org/10.48550/arxiv.2409.13714","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2409.13714","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:2409.13714","doi":null,"is_oa":true,"landing_page_url":"http://arxiv.org/abs/2409.13714","pdf_url":"https://arxiv.org/pdf/2409.13714","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":[],"has_content":{"pdf":true,"grobid_xml":true},"content_urls":{"pdf":"https://content.openalex.org/works/W4403752327.pdf","grobid_xml":"https://content.openalex.org/works/W4403752327.grobid-xml"},"referenced_works_count":0,"referenced_works":[],"related_works":["https://openalex.org/W7133187725","https://openalex.org/W7128544229","https://openalex.org/W4412888841","https://openalex.org/W4413682992","https://openalex.org/W7133185117","https://openalex.org/W6891983490","https://openalex.org/W7163604764","https://openalex.org/W4399038319","https://openalex.org/W7114932545","https://openalex.org/W4393090174"],"abstract_inverted_index":{"Achieving":[0],"a":[1,45,68,87,103,146,178],"mechanistic":[2],"understanding":[3],"of":[4,18,23,38,71,107,128,157,165],"transformer-based":[5],"language":[6,97],"models":[7,72,98],"is":[8,78],"an":[9],"open":[10],"challenge,":[11],"especially":[12],"due":[13],"to":[14,61,131,175],"their":[15,31,117],"large":[16,69,96],"number":[17,70],"parameters.":[19],"Moreover,":[20],"the":[21,35,126,162,166],"lack":[22],"ground":[24,53],"truth":[25,54],"mappings":[26,55],"between":[27],"model":[28],"weights":[29],"and":[30,80,100,111,116,136,149,183],"functional":[32],"roles":[33],"hinders":[34],"effective":[36],"evaluation":[37],"interpretability":[39,76,92,185],"methods,":[40],"impeding":[41],"overall":[42],"progress.":[43],"Tracr,":[44],"method":[46],"for":[47,74,90,181],"generating":[48,91],"compiled":[49],"transformers":[50],"with":[51,145],"inherent":[52],"in":[56],"RASP,":[57],"has":[58],"been":[59],"proposed":[60],"address":[62],"this":[63,83,122,139],"issue.":[64],"However,":[65],"manually":[66,109],"creating":[67],"needed":[73],"verifying":[75],"methods":[77],"labour-intensive":[79],"time-consuming.":[81],"In":[82],"work,":[84],"we":[85,124],"present":[86],"novel":[88,104],"approach":[89],"test":[93,159],"beds":[94],"using":[95],"(LLMs)":[99],"introduce":[101],"TracrBench,":[102],"dataset":[105],"consisting":[106],"121":[108,171],"written":[110],"LLM-generated,":[112],"human-validated":[113],"RASP":[114,134],"programs":[115,135],"corresponding":[118],"transformer":[119],"weights.":[120],"During":[121],"process,":[123],"evaluate":[125],"ability":[127],"frontier":[129],"LLMs":[130],"autonomously":[132],"generate":[133],"find":[137],"that":[138],"task":[140],"poses":[141],"significant":[142],"challenges.":[143],"GPT-4-turbo,":[144],"20-shot":[147],"prompt":[148],"best-of-5":[150],"sampling,":[151],"correctly":[152],"implements":[153],"only":[154],"57":[155],"out":[156],"101":[158],"programs,":[160],"necessitating":[161],"manual":[163],"implementation":[164],"remaining":[167],"programs.":[168],"With":[169],"its":[170],"samples,":[172],"TracrBench":[173],"aims":[174],"serve":[176],"as":[177],"valuable":[179],"testbed":[180],"evaluating":[182],"comparing":[184],"methods.":[186]},"counts_by_year":[],"updated_date":"2026-10-06T10:28:23.643775","created_date":"2024-10-25T00:00:00"}