Paper accepted at ICLR: a fun collaborative paper with the Polack lab (Rutgers University), featuring Julien Corbo, Batuhan Erkat, and Pierre-Olivier Polack, about biases emerging during task learning. The theoretical part was spearheaded by Keith Murray (now at Princeton) and Balázs Meszéna, and joined by Marci Hajnal. Contextual learning with a taylor made VAE, the Task-Amortized VAE is introduced.