Terrance Tao on AI vs. traditional software

Here’s an interesting observation:

“With enough experience reading mathematical texts, one tends to pick up a number of the contextual and stylistic clues that can lead one to evaluate the validity, depth, and importance of any given portion of the argument. This allows for much quicker reading, and isolating the "meat" of a paper.

But these signals are barely useful for reading #AI-generated mathematics - now texts look convincing but only slowly reading the text line by line can reveal the flaws...”

More here: https://mathstodon.xyz/@tao

And specifically: https://mathstodon.xyz/@tao/109945628011027107

This explains why "AI" definitely needs human oversight and control, especially at thr current stage.

Well put but obvious? Am I missing the rest of a thread?