RL for closed-loop adaptive BCIs and sim-to-real robotics still look like the strongest practical directions right now.