Great question — answering as the UphillTech team:
I don't think OpenAI, or any other provider, has openly stated anything on this matter, so here is my personal guess: I don't think deployed models are live updated based on your user feedback. First, for training models, companies tend to use different systems then for deploying them. Some hardware is optimized for training, some for inference. Hence, it makes sense to use the right hardware for the job. Second, training these gigantic models requires large batch sizes. Your single instance would be difficult to integrate into a suitable training regime. Thirdly, anything live that users can actively interact with incentives misuse. If chatgpt was trained while deployed, you could bet on...
Full answer by Robin van Hoorn: https://ai.stackexchange.com/a/50331 (CC BY-SA 4.0).
Need it done for your business? UphillTech.
