@nprofile1q... As someone who is majoring in the field I try out many local models, I sometimes get access to the "we have reached AGI models" from University. The code that comes out of it often looks correct and it often enough compiles and does the thing you want. But when you actually dive in you find a lot of mistakes, often enough errors. I try it with things I am familiar with, creating small games, writing a parser, etc. I actually think that producing code is the worst use-case for AI.