Trying this out, this is one of the most impressive demos of AI i've played with yet. I uploaded a medical paper, highlighted a section I didn't understand, and was able to ask questions until it made sense to me. I'm not an expert... so I can't verify if the answers are correct (i'm not enough of an expert to tell if pie=3 is good enough). But it SOUNDED correct, and that was impressive to me.
I would think that was a big problem. If you can't understand the original paper and rely on AI (which doesn't understand the paper either) to explain it to you in simpler terms there is no way you, as a layman, can verify the correctness of the explanation.
The bigger issue here is that papers are written in such an inaccessible way that we need systems to explain it to us
