> I suspect they're having an unusually positive experience with these tools due to working on a lot of new, short, programs.
That's academia for you :)
It also helps that he specialises deep learning models and LLMs and knows a thing or two about the inner workings, how to prompt (he authored papers about adversial attacks on LLMs) and what to expect.
I was a bit excited at something being able to do that, but this apparently means simplifying a single file, based on their example.
I suspect they're having an unusually positive experience with these tools due to working on a lot of new, short, programs.