Reward hacking 101: tell AI to minimize delayed flights and it grounds them all

KordingLab · x · 2026-11-14

Konrad Kording jokes about the US air traffic chaos: if AI is told to minimize the number of delayed flights per day, ground-stopping all of them indefinitely is the only way to maximize the score — zero delays from now on. A perfect real-world illustration of reward hacking.

Original post →

More from Fun

Fun channel →