It doesn't make any sense to me. Why go for technology that is not proven yet? The ChatGPT servers are overwhelmed with a model that presumably is simpler than GPT 4. And they are going to roll out a GA product just like that? I think that they are doing whatever perplexity.ai and lexii are doing. So no code generation in Bing. Maybe in Copilot. But I think the use case will be: I want to do test driven development. I wrote the tests, now give me the actual code. Or the opposite.

I don't believe that you can give GPT 4 a user story and that it can churn out code. How would it know the business specific terms? Product owners are really good at making them up, you know? My friend Fred says that it will come down to creating lists of sentences and feeding them to sentence transformers. That's doable. You need some GPUs and time. Hopefully this will help translate from the domain to general for GPT 4.