What's being ignored in your idea is that ChatGPT itself doesn't consist of code. It's a large artificial neural network, trained on huge amounts of data. It's not an algorithm. The solution of the problem of making it better is not to throw more coding power onto it. Instead you need better concepts, more/better training data, huge amounts of computing time, more human feedback. You can't just take a "chunk of code" of the network and "improve it".