Is Reactor Learning the Best Way to Train Your Agents?

Reinforcement learning gets very few bits of information because it only gets the rewards you need more supervision. Instead, like self supervised learning, which has been shown toi extract very rich information from the wold around you. For example, you can take a large web data set, and unsupervised, like, contrast of learning, gets you performance that not near as good as supervised learning but pretty close. It's actually better an like many task specific glosses you can do. Or you can use it till ie transfer to a variety of downto individual tasks.