Can you rescind an instruction? Eg you later decide you don’t want emojis. If you say, “stop using emojis” will it add a new instruction or will it know that it can remove the previous instruction to achieve the same effect?
the description mentions that "then automatically use that feedback to improve over time. It does this by creating few-shot examples from that feedback and incorporating those into the prompt."
@@crotonium I think that could be solved easily retrieving top-k last messages in a vector database or retrieving the last examples in a sql/nosql database to remember, but I am not sure.
My guess is it takes the top user scored items and then relays those if a related text is given to the agent. "Use these examples for a response that is desirable" etc.
Thanks for the video. May I know if the token will be increased as user provides more feedback? I am not sure. But I assume the feedback from the user is also fed into the prompt to change the behavior of the LLM?
