It is just plain old supervised learning. A regression from the post features to vote count. The RL discussion in TFA is a bit confusing. Such a model can be u

It is just plain old supervised learning. A regression from the post features to vote count. The RL discussion in TFA is a bit confusing. Such a model can be used as the "reward model" for the "reinforcement learning from human feedback" (RLHF) method.