> An inaccurate engine and a statistical one are indistinguishable. How so? The paper is trying to assess if some semantic understanding is being created by th

> An inaccurate engine and a statistical one are indistinguishable. How so? The paper is trying to assess if some semantic understanding is being created by the underlying model, not if a completely accurate one is. If such a world model maps to Othello-prime (gleaning concepts like tiles, colors etc), that is still a very interesting result.