Comment by throw10920

I have basic understanding about how refusal-removal works - find the "no" weights by intentionally generating diverse refusals, and then set those weights to zero.

Is there a similar process for removing not refusals, but misinformation?