It's unclear where in the paper they demonstrate, unequivocally, that reasoning occurred, rather than pure statistical model-fitting (we generally believe this is one of the key differences between AGI and modern LLMs)
> reasoning occurred, rather than pure statistical model-fitting
What's the difference?
reasoning takes into account much more contextual information. I think many would also say it's a step-by-step process in which the explicit assumptions and conclusions could be stated clearly, although I think that's probably just anthropomorphism and projection.
It gives me the correct answer (as judged by experts) to a complex medical problem. If you want to keep saying this isn't reasoning, be my guest. But IMO that's irrelevant.
Spitting out the results of a bunch of experts is the opposite of reasoning. Or more correctly, it's one limited form of reasoning: an ensemble prediction made from a mixture of experts. That's not an interesting form of reasoning because literally all the decisions are encoded in the weights you apply to the experts.
You seem very confused and I invite you to think again about what you said.