When people think about using GPT-3 for real products, I always wonder about the running costs. These models are so big that even inference can be fairly costly to run. Any thoughts on that?
Davinci is expensive, but the other models can be quite affordable even with finetuning. I dont see problems for a personal assistant based on them (edit: from a personal usage point-of-view. if you are making a product for others to use, that's different :)
