Re: bad with large code -- that's been something I've been noticing as well. A project I've been working on for work has gotten larger now and it seems more often that it makes goofy mistakes. Like removing something somewhere but leaving something related somewhere else. My best guess is that it's all context window stuff. As humans, we can learn a lot and keep it all in our heads. But an AI model, if it's not something from its training set and it's not in its context window, then it just straight up does not know it exists. I'm guessing because context windows have gotten bigger that ai coding has gotten more impressive. But we'll see if they're able to keep scaling that. I imagine at some point it's gonna hit a soft wall (or at least a price wall lol).