Comment by zjp

10 months ago

I occasionally pair write with LLMs, but I give them my piece and then say, "I don't want your edits, just your feedback" and ask them some simple questions about the content and different angles on it. When the LLM says what I want it to say, I consider the piece good enough. That is to say, if a machine understands what you're saying and a human doesn't, that human's criticism might be below engaging with.

>That is to say, if a machine understands what you're saying and a human doesn't, that human's criticism might be below engaging with.

No wonder OpenAI accidentally made LLMs even more sycophantic if this is what people want from them.