I'm surprised to see "mathematicians" here, but what do they refer to? Mathematician as an academic position? As teacher? As consultant?
And, more to the point, mathematicians are often doing that work because they enjoy it. Even in the best case for GPT, with respect to most professional mathematicians, you've done the equivalent of automating away their Sudoku/Crossword time.
You might be able to replace some graduate student labor or post-doc labor, but a huge reason for doing that labor in the first place is to train new mathematicians... it's like the grad school version of automating a second grader's multiplication table: you can do it, and you are "automating" the second grader's labor, but you've kinda missed the point...
I wonder how many of the other categories are similar, where the tasks that are being automated are sort of pointless or even counter-productive to automate. Because automating those tasks disrupts the learning process between the professional and their customer, and that learning process is actually the primary product. Legal secretaries and accountants in particular come to mind. Even a lot of web dev work.
This is absurd.
It seems AI is looking for problem, where better and cheaper options would make infinitely more sense...