It might. If hallucinations are truly random and not correlated to anything shared between models. For example, something inherent to the data they are trained on. Given how locked down I think potential training data is going to become, and the amount of data required, I think that sharing data between models is almost guaranteed.

Also that sounds like an awful lot of computing power for everyday UIs. It also doesn’t solve the non determinism problem.