Learning from human preferences

In three linesOpenAI and DeepMind develop a preference learning algorithm to infer human objectives without explicit reward functions, reducing risks of undesirable AI behaviors.

Summary generated by Claude — human-verified