[JumpCrisscross]

> How true is this? From playing around with Bard and Claude, GPT-4 seems to be significantly better

I have at most moderate confidence in this hypothesis.