New paper: task design, evaluation and models for 'attributed QA'. Here, we pose a question to an LM, and require that it produces an answer, paired with an 'attribution' in the form of a paragraph from a trusted index, such that the attribution supports the answer. 1/