[rogerrogerr]

I am 99% in favor of this, but the one place where I’d rather have the prompt _and_ the output is when the model had access to additional context that can’t be shared for some reason.

For example, if I’ve filed a support ticket with some software company, and they reply to me by prompting some LLM to “adapt the instructions from when we solved this with Acme Corp”, the prompt doesn’t do me much good without the (presumably confidential) correspondence with Acme.

I’d always rather read human written text, though.

[relaxing]

Just send me a link to the Acme Corp knowledgebase then.