The mechanisms of in-context learning in transformers are not currently well understood. Visit the #ICML2023 Google Research booth today at 3:30 PM for a demonstration on how a type of gradient descent enables transformers to pay better attention to their context.