Comment by whazor

This direction could be an interesting AI benchmark. All kinds of different humans use LLMs for their job, whether allowed or not. Including diplomats, defence personnel, lawyers etc etc. Within the benchmark you could play both sides and reward when both sides reach some kind of mutually beneficial game theory scenario where both parties win.