Comment by great_psy

Shouldn’t we expect a convergence as models become more powerful ?

There is an optimal answer to any question. Something that maximizes utility and minimizes tokens.