A standard advice would be: do what you are good at and enjoy doing, and keep an eye on the potential applications to LW-style rationality, should you find it interesting (or post-rationality, as is the fashion). For example, working on multi-agent models of agency/self can probably yield a lot of low-hanging fruit, especially if you do your research on the margins (like people whose sense of self is different from average, outward manifestations of self-ness in ML, etc.). You don’t have to get math-heavy and still figure out a lot of new insights.
A standard advice would be: do what you are good at and enjoy doing, and keep an eye on the potential applications to LW-style rationality, should you find it interesting (or post-rationality, as is the fashion). For example, working on multi-agent models of agency/self can probably yield a lot of low-hanging fruit, especially if you do your research on the margins (like people whose sense of self is different from average, outward manifestations of self-ness in ML, etc.). You don’t have to get math-heavy and still figure out a lot of new insights.