It doesn’t follow that because a system is Turing complete the approach being used will eventually achieve reasoning.

No, but that was also not the claim I made.

The point is that as the person I replied to pointed out, that LLM's are "next token predictors" is a meaningless dismissal, as they can be both next token predictors and Turing complete, and given that unless reasoning requires functions outside the Turing computable (we know of no way of constructing such functions, or no way for them to exist) calling them "next token predictors" says nothing about their capabilities.