Pattern comments on Value Learning is only Asymptotically Safe

[Pattern]

is there anything to aim for that is stronger than asymptotic safety?

Faster convergence?