One would hope, that since the problem these models are trying to solve is language modeling, they would eventually converge around similar capabilities