I think in this case, people tend to underrate just how capable and flexible the basic LLM architecture is. And, also, underrate how many gains are there in bet

I think in this case, people tend to underrate just how capable and flexible the basic LLM architecture is. And, also, underrate how many gains are there in better training vs better architecture.