Scientists Use Human Preferences to Train AI Agents 30x Faster | HackerNoonThe study evaluates a method through two experimental approaches: proxy human preferences and real human preferences.