Great question! Back when we initiated Diplomacy, I hypothesized that "an agent that can read the rules of any game and play it at an intermediate level" would be the next challenge problem. There's been so much progress on the language modeling side that I think a system like this is within reach within the next 2-3 years if substantial effort was devoted to it. We're starting to see similar task-generality in large language models on real-world tasks, although constructing a symbolic representation for planning out of a text description is still an open research question! -AL