The benchmark shows how various LLM-produced texts are shifted in style compared to human-written texts. Details are in this paper by Jiří Milička, Anna Marklová, and Václav Cvrček. Based on Biber's classic MDA stylistic vectors (Biber 1988) and their adaptation for Czech by Cvrček et al. (2018). The LLM-generated corpora (AI Brown v2 and AI Koditex v2) are searchable via the Kontext interface (Universal Dependencies tagging). Click a heatmap cell to open a scatterplot (every dot is one text chunk).