Comment by evanb

I'm a professor of physics; my research area is computational quantum field theory. I was beginning a new approach for a certain kind of model. I know the model has not been simulated before (a collaborator has just constructed it for the first time using in-field knowledge developed in the last year).

I've been steadily more impressed, but the real startling moment was earlier this week, I could show that in a particular operation I had a bug (some identity wasn't satisfied) but I couldn't understand why. In one try Claude Code grokked my code, explained the problem to me (there were some dimension-dependent minus signs that needed to be computed), and fixed it.