Can a 9B open model learn to plan robot missions from temporal-logic specifications as well as a much larger one? Here’s the story of how we got there.

Why should a model that fits a finite set of training examples do well on data it has never seen? Here’s a gentle tour of how the PAC framework answers that.