I have already built a system that puts ChatGPT in a loop with the output of programs and running commands. It can be pretty effective. It's even more effecti

I have already built a system that puts ChatGPT in a loop with the output of programs and running commands. It can be pretty effective. It's even more effective when you can limit the domain somewhat to things that can fit in the context window, which eliminates out of date API information. We will now shortly be able to add visual feedback of the applications into the loop.