we are still "early" in a lot of the algorithmic and architectural progress available in transformer models more good evidence of this, no free lunch but certainly a lot of opportunities out there.