You curiously leave out a significant sentence in his post, "After two weeks of further work, we were able to check everything and put the prose in a form that could be understood by mere mortals."

This is not a model one-shotting a published paper! It's not even published! Anyone with a name can get a preprint up. I read this to mean that he gave it a prompt to find problems, then _he evaluated and chose the problem_, undoubtedly to be someone simple and quick which could be presented in less than 5 pages. After that he let the program work on its own for "a couple of hours". After which he spent _2 weeks_ with the model to improve the results. This does not argue against my point, this is exactly what I'm saying. Of course a model can do parts of the process, but it's terrible at writing it up "in a way mere mortals can understand". You have to hand hold and guide and do work yourself. And this is by an established professor in the field with a record of publications.

I ask you again, have _you_ used this technology to perform graduate+ level research?