Comment by 99112000

Breaking News, BuzzFeed man can take a joke amd fires back.

I appreciate defining a clear hypothesis and the exploring an LLM using statistics. I feel like the analysis could benefit from prompts that contain neutral consequenses as well. You have given it clear positive rewards, clear negative ones and no reward. Neutral consequences may be a better baseline than no reward.