Why reinforcement learning plateaus without representation depth (NeurIPS 2025)

Wow, they finally figured out that it's actually not A, it's B—at least if C.