Gathering human feedback

In three linesOpenAI releases RL-Teacher, an open-source implementation for training AIs via occasional human feedback instead of hand-crafted reward functions. The technique aims to develop safe AI systems and applies to reinforcement learning problems where rewards are hard to specify.

Summary generated by Claude — human-verified