Comment by volkercraig

I'm not saying you can't compare them, I'm saying it's pointless. LLM's are extremely large scale multivariate regression machines, evaluating it's output within it's own training domain is as pointless as seeing if a ball rolls downhill.