How academia is adapting to ChatGPT