Reinforcement learning

Deep down I am a control/RL junkie, but I’ll give it to the NLP crowd, they have managed to turn AI mainstream and create a useful tool. It’s not that we are not seeing really cool RL projects (see here: https://clemenswinter.com/2023/04/14/entity-based-reinforcement-learning/, here: https://github.com/NVIDIA-Omniverse/IsaacGymEnvs and here: https://arxiv.org/abs/2304.01315), it’s that they do not have much use outside the lab (yet).

In a sense, we still do not have the killer RL application, and as I have said in another post, we will not get one unless zero-shot RL is solved.