I suspected this would happen. At the end of the day larger models have more to work with, this makes a big difference. Also, there's a lot of domain knowledge

I suspected this would happen. At the end of the day larger models have more to work with, this makes a big difference. Also, there's a lot of domain knowledge in GPT-4 which isn't code but surely makes a big difference when it comes to understanding what you want, and the context of the problem and the solution.