I think this is a poor use case for GPT-3, although since we’re so early on that I suppose any use case is better than none. If the code has an error, you already need to be skilled in Power Fx to debug, and given my own experience trying to get GPT-3 to create CSS animations, that would be frequent. Especially assuming Microsoft has much wider access to re-training the algorithm to specialized use cases, expanded prompts, and anything else that comes with their owning the source code, I think they could be a lot less boring than this.
1 Like
I wonder if this will be integrated into DevOps? There’s already a lot of ML and automation in DevOps pipelines today. It will probably be better at fixing bugs than writing code from scratch for a while.