GATO is able to retain knowledge across different tasks. It’s not, however, is able to "generalize"(i.e. the improvement in one task did not lead to improvement in a different task) (if i remember the paper correctly). So no, AGI was not solved.