2026-10-11T12:37:38Z https://api.figshare.com/v2/oai

oai:figshare.com:article/33814144 2026-09-15T17:55:55Z category_734 category_931 category_106 portal_5 item_type_3 month_year_09_2026

<p>Zero-shot accuracy of four evaluated large language models across seven regulatory functional subdomains.</p> Xuejing Fu (24950107) Zhen Chen (129176) Wentao Lu (7905074) Biological Sciences not elsewhere classified Information Systems not elsewhere classified Science Policy stratified independent review stable dataset release jurisdictional pharmaceutical regulation jurisdiction question answering 42 – 8 autonomous regulatory interpretation support workflows rather observed residual material overall accuracy ranged two regulatory experts systematically evaluated using div >< p 29 %– 5 regulatory applications two highest material errors findings support monolingual accuracy wilson 95 supervised decision subsequently classified standardized zero shot protocol semantic screen scoring models results indicate reproducible resource negative population multiple comparisons max — increasingly explored including 10 evaluation resources error proportion drug administration differ significantly consistently lower confidence interval bilingual multiple absolute differences 62 %). 444 items 432 items 43 %. 122 classified <p>Zero-shot accuracy of four evaluated large language models across seven regulatory functional subdomains.</p> 2026-09-15T17:55:41Z Dataset Dataset 10.1371/journal.pone.0343858.t007 https://figshare.com/articles/dataset/_p_Zero-shot_accuracy_of_four_evaluated_large_language_models_across_seven_regulatory_functional_subdomains_p_/33814144 CC BY 4.0