I agree. Coming up with the right prompts was not trivial. I almost quit several times. Yet, there is a science to this and I think it’ll become more important to turn out focus away from the spectacle aspects of GPT and more towards reproducibility. More so if the way forward is via interrelated instances of GPT.
As an aside, critique seems much easier than generation. I’m cautiously optimistic about prompting GPT instances to “check” output.
I agree. Coming up with the right prompts was not trivial. I almost quit several times. Yet, there is a science to this and I think it’ll become more important to turn out focus away from the spectacle aspects of GPT and more towards reproducibility. More so if the way forward is via interrelated instances of GPT.
As an aside, critique seems much easier than generation. I’m cautiously optimistic about prompting GPT instances to “check” output.
Similar to sharp google-fu today but much deeper.