Terrible take. Fraud was already hard to catch, partly because it was hard to look for. Generative AI means that the actual data for every paper needs to be s

Terrible take. Fraud was already hard to catch, partly because it was hard to look for. Generative AI means that the actual data for every paper needs to be scrutinized and tested in exactly the same way it always should have been but wasn’t. This is like saying that the internet made computer security worse. No, it didn’t. It exposed all sorts of terrible practices that were never secure but which were good enough in a high trust model. Moving to low/zero trust has been good for security, and it will be good for science. It’s just going to be a painful decade or two as past bad practices fail spectacularly rather than subtly.