[phreeza]

The obvious follow up question: are there methods in use to bait such models into suboptimal play and then switch play style to exploit that?