Hey, really cool post! These accuracy numbers aren’t conditional on controllability task success, right? How is the accuracy conditional on controllability success?
Julie Steele
Karma: 0
Hey, really cool post! These accuracy numbers aren’t conditional on controllability task success, right? How is the accuracy conditional on controllability success?
Another question—why did you use the user message for the CoT replacement but a system prompt for the CoT skipping?