Comment on Task-free intelligence testing of LLMsparentComments−esafak8moDoesn't that presume that one model dominates the other?−nestorD8moIt presumes some models are better than others (and we do find that providing data with a wide mix of model strengths improves convergence) but it does not need to be one model, and it does not even need to be transitive.
Comments
Doesn't that presume that one model dominates the other?
It presumes some models are better than others (and we do find that providing data with a wide mix of model strengths improves convergence) but it does not need to be one model, and it does not even need to be transitive.