Contrastive Learning as Goal-Conditioned Reinforcement Learning

Did really good in some of the OGbench tasks as shown in the Horizon Reduction Makes RL Scalable.