AI"s inability to continually learn remains one of the biggest problems standing in the way to truly general purpose models. Might it soon be solved?