Comment on Using reinforcement learning and $4.80 of GPU time to find the best HN postparentComments−jampekka1yIt 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.
Comments
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.