It is interesting to think about how to decide what to work on, especially in the age of AI. I think one thing that’s becoming clear to me is that I want to work on something that will be aided by more capable AI systems, and that will not be washed away by AI itself. This might sound obvious, but in practice, many things do not satisfy both constraints. Some questions that can help assess this are: is the usefulness of what I’m working on invariant to increasingly performant AI models being available? If new AI systems are announced or launched, will I feel nervous/sad, or will I eagerly anticipate the release of these systems?