Why does it have to be an either-or? Maybe LLMs are doing a bit of both, in a weird hybrid way; it is both doing a statistical calculation and yet the network i

Why does it have to be an either-or? Maybe LLMs are doing a bit of both, in a weird hybrid way; it is both doing a statistical calculation and yet the network is parameterized to do some very rudimentary (and nonhuman) reasoning computations. That's plausible to me, and explains why the reasoning is so hard to isolate... Just like how looking at a human brain it is hard to isolate the reasoning capacities.