Comment by gchamonlive

Only if you think in terms of perceived raw intelligence, but self-update is a form of valuable self-improvement that could benefit current models a lot, if they could commit facts from context into their weights cheaply and reliably.