Hacker News Reader

It reminds me of the thread about the Quake 3 bots, who left alone for several years, figured out that the best approach was to not kill each other.

Without knowledge of their reward function its difficult to tell if they're converged on this strategy or if its just broken.