New paper! We trained reasoning models to reason about their uncertainty using RL!