How do you guys get the model to generate the code you want that actually works. And building off of that code generating more code which is compatible?
I have been working on some things like this, trying to prompt engineer my way to victory. It’s going well using 3.5 but 4 would be great, and the plugins would be great too. I am thinking I mainly need to get my prompts locked in.
Anyone else have similar or other approaches?
linus
2
Hi @Harrison1,
welcome to the forum 
Can you share some details on what you were trying to do, then we can provide better answers.
In general I’d advice to not use the 3.5 or 4 models but the models dedicated for code generation! This helps A LOT! So either going to Github and using the solution there (based on OpenAI) or go to the playground ( Playground - OpenAI API) and then selecting a codex-model. Those models are trained with better data regarding the coding usecase and perform a lot better than the general models!
