For How do you improve evaluation when working with AI Red Teaming?, I would separate the problem into setup, output quality and repeatability. evaluation and working fits into that because it can change the result quite a lot depending on the workflow. If you share the exact setup, it would be...