The problem with this line of thinking is that you can then only really trust your own personal bubble. Or actual trial and error, which is costly.

These models get ever better at producing plausible text. Once they permeate the academia completely, we're cooked.

And even academia is not clean for some matters, or complete.