Tripping Up AI

Tomer Ullman discusses his experiments with tripping up AI models by introducing task variations and reveals the challenges in accurately reporting the results.