I really like that post, but I view it as a description of the limit case of how an arbitrarily powerful LLM would behave. It doesn't give you a way to compare models or predict their limitations or relative advantages.