{"id":"https://openalex.org/W4394642192","doi":"https://doi.org/10.48550/arxiv.2404.05399","title":"SafetyPrompts: a Systematic Review of Open Datasets for Evaluating and Improving Large Language Model Safety","display_name":"SafetyPrompts: a Systematic Review of Open Datasets for Evaluating and Improving Large Language Model Safety","publication_year":2024,"publication_date":"2024-04-08","ids":{"openalex":"https://openalex.org/W4394642192","doi":"https://doi.org/10.48550/arxiv.2404.05399"},"language":"en","primary_location":{"id":"pmh:oai:arXiv.org:2404.05399","doi":null,"is_oa":true,"landing_page_url":"http://arxiv.org/abs/2404.05399","pdf_url":"https://arxiv.org/pdf/2404.05399","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":"review","indexed_in":["arxiv","datacite"],"open_access":{"is_oa":true,"oa_status":"green","oa_url":"https://arxiv.org/pdf/2404.05399","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5050023117","display_name":"Paul R\u00f6ttger","orcid":"https://orcid.org/0009-0008-7115-6893","observed_orcids":["https://orcid.org/0009-0008-7115-6893"]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"R\u00f6ttger, Paul","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5095761893","display_name":"Fabio Pernisi","orcid":null,"observed_orcids":[]},"institutions":[{"id":"https://openalex.org/I71209653","display_name":"Bocconi University","ror":"https://ror.org/05crjpb27","country_code":"IT","type":"education","lineage":["https://openalex.org/I71209653"]}],"countries":["IT"],"is_corresponding":false,"raw_author_name":"Pernisi, Fabio","raw_affiliation_strings":["Bocconi University ,"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Bocconi University ,","institution_ids":["https://openalex.org/I71209653"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5084936290","display_name":"Bertie Vidgen","orcid":"https://orcid.org/0000-0002-7892-0814","observed_orcids":["https://orcid.org/0000-0002-7892-0814"]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Vidgen, Bertie","raw_affiliation_strings":["Contextual AI"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Contextual AI","institution_ids":[]}]},{"author_position":"last","author":{"id":"https://openalex.org/A5084505122","display_name":"Dirk Hovy","orcid":"https://orcid.org/0000-0002-4618-3127","observed_orcids":["https://orcid.org/0000-0002-4618-3127"]},"institutions":[{"id":"https://openalex.org/I71209653","display_name":"Bocconi University","ror":"https://ror.org/05crjpb27","country_code":"IT","type":"education","lineage":["https://openalex.org/I71209653"]}],"countries":["IT"],"is_corresponding":false,"raw_author_name":"Hovy, Dirk","raw_affiliation_strings":["Bocconi University ,"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Bocconi University ,","institution_ids":["https://openalex.org/I71209653"]}]}],"institutions":[],"countries_distinct_count":1,"institutions_distinct_count":1,"corresponding_author_ids":[],"corresponding_institution_ids":[],"apc_list":null,"apc_paid":null,"fwci":0.3054,"has_fulltext":true,"cited_by_count":4,"citation_normalized_percentile":{"value":0.5041876,"is_in_top_1_percent":false,"is_in_top_10_percent":false},"cited_by_percentile_year":{"min":94,"max":96},"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/T10883","display_name":"Ethics and Social Impacts of AI","score":0.6449999809265137,"subfield":{"id":"https://openalex.org/subfields/3311","display_name":"Safety Research"},"field":{"id":"https://openalex.org/fields/33","display_name":"Social Sciences"},"domain":{"id":"https://openalex.org/domains/2","display_name":"Social Sciences"}},"topics":[{"id":"https://openalex.org/T10883","display_name":"Ethics and Social Impacts of AI","score":0.6449999809265137,"subfield":{"id":"https://openalex.org/subfields/3311","display_name":"Safety Research"},"field":{"id":"https://openalex.org/fields/33","display_name":"Social Sciences"},"domain":{"id":"https://openalex.org/domains/2","display_name":"Social Sciences"}},{"id":"https://openalex.org/T11689","display_name":"Adversarial Robustness in Machine Learning","score":0.16539999842643738,"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/T12262","display_name":"Hate Speech and Cyberbullying Detection","score":0.1264999955892563,"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-model-safety","display_name":"large language model safety","score":0.9380000233650208},{"id":"https://openalex.org/keywords/dataset-evaluation","display_name":"dataset evaluation","score":0.5260000228881836},{"id":"https://openalex.org/keywords/synthetic-dataset","display_name":"synthetic dataset","score":0.800000011920929},{"id":"https://openalex.org/keywords/catastrophic-risk","display_name":"catastrophic risk","score":0.925000011920929},{"id":"https://openalex.org/keywords/llm-benchmark","display_name":"LLM benchmark","score":0.7480000257492065}],"concepts":[{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.5292059183120728},{"id":"https://openalex.org/C2522767166","wikidata":"https://www.wikidata.org/wiki/Q2374463","display_name":"Data science","level":1,"score":0.34585487842559814},{"id":"https://openalex.org/C204321447","wikidata":"https://www.wikidata.org/wiki/Q30642","display_name":"Natural language processing","level":1,"score":0.33735191822052}],"mesh":[],"locations_count":2,"locations":[{"id":"pmh:oai:arXiv.org:2404.05399","doi":null,"is_oa":true,"landing_page_url":"http://arxiv.org/abs/2404.05399","pdf_url":"https://arxiv.org/pdf/2404.05399","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.2404.05399","doi":"https://doi.org/10.48550/arxiv.2404.05399","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2404.05399","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:2404.05399","doi":null,"is_oa":true,"landing_page_url":"http://arxiv.org/abs/2404.05399","pdf_url":"https://arxiv.org/pdf/2404.05399","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":[{"display_name":"Systematic Review","id":"https://openalex.org/study-designs/systematic-review"}],"awards":[{"id":"https://openalex.org/G434337789","display_name":null,"funder_award_id":"HORIZON2020","funder_id":"https://openalex.org/F4320320300","funder_display_name":"European Commission"},{"id":"https://openalex.org/G6530115474","display_name":"Incorporating Demographic Factors into Natural Language Processing Models","funder_award_id":"949944","funder_id":"https://openalex.org/F4320338335","funder_display_name":"H2020 European Research Council"}],"funders":[{"id":"https://openalex.org/F4320320300","display_name":"European Commission","ror":"https://ror.org/00k4n6c32"},{"id":"https://openalex.org/F4320338335","display_name":"H2020 European Research Council","ror":"https://ror.org/0472cxd90"}],"has_content":{"pdf":true,"grobid_xml":true},"content_urls":{"pdf":"https://content.openalex.org/works/W4394642192.pdf","grobid_xml":"https://content.openalex.org/works/W4394642192.grobid-xml"},"referenced_works_count":0,"referenced_works":[],"related_works":["https://openalex.org/W4386228564","https://openalex.org/W4402670430","https://openalex.org/W7139143845","https://openalex.org/W7119466076","https://openalex.org/W4386755570","https://openalex.org/W4414858165","https://openalex.org/W7126420743","https://openalex.org/W4414875577","https://openalex.org/W4404407088","https://openalex.org/W4406485177"],"abstract_inverted_index":{"The":[0],"last":[1],"two":[2],"years":[3],"have":[4,22],"seen":[5],"a":[6,111,150,165,207,219],"rapid":[7],"growth":[8],"in":[9,45,52,97,160,182,185],"concerns":[10,25],"around":[11,61],"the":[12,56,68,85,138,235],"safety":[13,178,239],"of":[14,30,40,58,70,115,140,168,205,210,222,237],"large":[15],"language":[16],"models":[17],"(LLMs).":[18],"Researchers":[19],"and":[20,34,47,63,81,93,120,134,146,170,189,202],"practitioners":[21,82],"met":[23],"these":[24,107],"by":[26],"creating":[27],"an":[28,132],"abundance":[29],"datasets":[31,88,117,179,224],"for":[32,79,89,118,225],"evaluating":[33,119],"improving":[35,121],"LLM":[36,122,177,186,191,226,238],"safety.":[37,123],"However,":[38],"much":[39],"this":[41],"work":[42,102],"has":[43],"happened":[44],"parallel,":[46],"with":[48],"very":[49],"different":[50],"goals":[51],"mind,":[53],"ranging":[54],"from":[55],"mitigation":[57],"near-term":[59],"risks":[60],"bias":[62],"toxic":[64],"content":[65],"generation":[66],"to":[67,83,94,231],"assessment":[69],"longer-term":[71],"catastrophic":[72],"risk":[73],"potential.":[74],"This":[75],"makes":[76],"it":[77],"difficult":[78],"researchers":[80],"find":[84],"most":[86],"relevant":[87],"their":[90],"use":[91,204],"case,":[92],"identify":[95],"gaps":[96,159],"dataset":[98,161],"coverage":[99],"that":[100,195],"future":[101],"may":[103],"fill.":[104],"To":[105],"remedy":[106],"issues,":[108],"we":[109,129,229],"conduct":[110],"first":[112],"systematic":[113],"review":[114,125],"open":[116,223],"We":[124,143,173],"144":[126],"datasets,":[127,155],"which":[128,228],"identified":[130],"through":[131],"iterative":[133],"community-driven":[135],"process":[136],"over":[137],"course":[139],"several":[141],"months.":[142],"highlight":[144],"patterns":[145],"trends,":[147],"such":[148,163],"as":[149,156,158,164,234],"trend":[151],"towards":[152],"fully":[153],"synthetic":[154],"well":[157],"coverage,":[162],"clear":[166],"lack":[167],"non-English":[169],"naturalistic":[171],"datasets.":[172,212],"also":[174],"examine":[175],"how":[176],"are":[180,199,215],"used":[181],"practice":[183],"--":[184,193],"release":[187],"publications":[188],"popular":[190],"benchmarks":[192],"finding":[194],"current":[196],"evaluation":[197],"practices":[198],"highly":[200],"idiosyncratic":[201],"make":[203],"only":[206],"small":[208],"fraction":[209],"available":[211],"Our":[213],"contributions":[214],"based":[216],"on":[217],"SafetyPrompts.com,":[218],"living":[220],"catalogue":[221],"safety,":[227],"plan":[230],"update":[232],"continuously":[233],"field":[236],"develops.":[240]},"counts_by_year":[{"year":2025,"cited_by_count":2},{"year":2024,"cited_by_count":2}],"updated_date":"2026-10-16T09:59:22.235173","created_date":"2024-04-10T00:00:00"}