systematicity of reasoning skills, and paves a path towards open-domain systems that constantly improve by interacting

In this work we focus on testing weather models can systematically reason over implicit knowledge. Below we answer all questions and provide results for requested additional experiments. R1: The quality and usefulness of the distractors is unclear. Thank you for your comment. Without distractors the models find biases in the data that hurt generalization.