Models inherit the writer, not who the writer was imitating

Research shows that language models trained on text inherit the style and characteristics of the original writer rather than the style of any imitated authors within the text. This finding clarifies how models capture authorial identity distinct from mimicry.