Request: API switches/sliders for behaviors like confabulate, creativity, etc

I don’t know if GPT-3 is even hypothetically capable of this right now, but perhaps future iterations could be. It would be really phenomenal if there we had a switch in the API that allowed us to enable or disable behaviors such as confabulation. That is to say if we could tell the engine to stick with cold hard facts (as in with information retrieval) or if we can tell it to make stuff up (creativity, fiction).

Perhaps this could be displayed as more sliders in the future similar to Temperature. Furthermore, it may be good if there were various endpoints for other common features like chat generation, as there are already endpoints for question answering.

GPT-3 is excellent because it has a lot of encyclopedic knowledge embedded BUT it can also just make stuff up on the fly as well, and we have no idea whether or not it is true.

Thanks for your feedback. At this time, the requested behaviour is achieved through prompt design. In particular, please see our documentation on factual responses. You can also use the Answers endpoint to query a set of documents, which is less likely to confabulate responses outside of what’s present in the documents.