Build, evaluate, keep what works, improve. Then do it again.
policy = init("curiosity") while True: data = rollout(policy, world) # act reward = evaluate(data) # real-world test policy = improve(policy, data, reward) log("iteration", )
$ git tag -n --sort=-date # news
$ git log --graph # research (left lane) and path (right lane) · * equal contribution · † corresponding
$ git show cv