The Programmers Teaching Machines to Forget
As models remember more, a small field is asking a harder question: how do you make one reliably unlearn?
Deleting a fact from a trained model turns out to be far harder than adding one. A handful of labs think that problem is the next real frontier.