My supervisor's wife was working at a pharma company back in the 2000s. Her job was to reproduce promising publications related to any conditions they were involved in. The reproducibility rate was something like 25%, which is higher than some other estimates I've seen looking across many fields, but still....

Incentives matter and right now they're the wrong ones

>Nope, the PR was poorly written (maybe the Department is experimenting with chatGPT ?).

ChatGPT probably would have done a much better job with access to the publication. Pretty soon, these lousy science journalists are all going to be out of a job when they can't even get basic facts correct, and the real scientists don't have time to write or review PR articles themselves, so an AI will fill that role instead.