Data and code: taxonomic distance of species confusions by large language models

Abstract

Data and code for the article Borrowed from a relative: how close the lending species is when a large language model attributes another species' record (Ji Sun Hong and Yuyong Kim). The article asks whether the species whose record a large language model borrows, when it answers about another species, is taxonomically structured. It reanalyses archived model responses about Korean protected species from three companion studies (no new model output) and measures the distance between the species asked about and the lending species on the GBIF Backbone Taxonomy (number of shared ranks) and on the Open Tree of Life synthetic tree, against random-register and permutation nulls. The archive contains the Korea Heritage Service Natural Monument register snapshot (27 September 2026), GBIF Backbone lineages, Open Tree of Life API responses, every result table and figure reported in the article, the numbers printed in the article, and the analysis scripts. The core statistics and figures can be reproduced from the archive alone (see README.md). Data are released under CC BY 4.0 and code under the MIT License.