Our first paper, investigating the formation of world models and simple path-following circuits in maze-solving transformers, has been accepted to the UniReps workshop at NeurIPS23! The article and accompanying blog post will be made public in the week following November 16th.