It is possible to run the output automatically and feed the errors back into it. I am am doing this. With the right prompting and feedback it is starting to wor

It is possible to run the output automatically and feed the errors back into it. I am am doing this. With the right prompting and feedback it is starting to work for short straightforward functions some of the time. It helps a lot to give it the test cases beforehand. I am hoping I can launch it as a service when they have a ChatGPT model in the API. For now using the 'unofficial' API.