AIS is an evaluation framework for assessing whether the output of natural language models only contains information about the external world that is verifiable in source documents, or "Attributable to Identified Sources". - View it on GitHub