Natural language replaces hand-coded reward functions in AI agent training

Original: Karpathy's prediction about RL is coming true now!

Source: x.com ↗