I wanted a more accessible way to play with GPT3, and as I spend a lot of my time in the terminal, this was a fun way! Check out the examples section to see some use cases.
Somewhat; I think I briefly mention this in the readme but I could expand on it. There is some reading here https://beta.openai.com/docs/guides/safety-best-practices and here https://beta.openai.com/docs/guides/production-best-practice..., although in general, yes, anything that gets sent to OpenAI is indeed, well, sent to OpenAI :). I wouldn't run any sensitive data through it for example, although I am not too concerned about my query history or anything like that (Write me a script for a hypothetical third Bladerunner film.) Also, similar to not curling into bash, I think it is unsafe to eval anything that comes out of this without first reading it.
