Philip Cohen, reflecting on the intersection of the perverse incentives of ‘publish or perish’ + low consequence editorial accountability + generative AI:

I assume you are appalled at the idea of journals publishing such fabricated results. However, there are a lot of people who would draw a different lesson from this example: Why spend an hour getting this real data, doing some recodes, running the model, and sharing the data and code — when ChatGPT already “knows” the answer? They see the small errors in the table above, and compare the time it took to produce them, and conclude that the ChatGPT approach is much more efficient and almost as good. This is productivity!