AI evaluation may bias perceptions: The importance of context in interpreting academic writing

In three linesStudy showing that evaluation methods for AI use in scientific publications produce significant biases when ignoring contextual differences across countries and fields. Pooled benchmarks conflate pre-existing stylistic variation with LLM-generated text, overestimating AI in some contexts and underestimating it in others.

Summary generated by Claude — human-verified