CSSI Research Seminar: Mohit Iyyer

Evaluating the factuality of LLM-generated text

Abstract: Recent advances in large language models (LLMs) have enabled them to process texts that are millions of words long, fueling demand for long-form language processing tasks such as the summarization or translation of books. However, LLMs struggle to take full advantage of the information within such long contexts, which contributes to factually incorrect text generation. In this talk, I consider the growing problem of long-form evaluation: as the length of the inputs and outputs of long-form tasks grows further, how do we even measure progress? I propose a high-level framework (applicable to both human and automatic evaluation) that first decomposes a long-form text into simpler atomic units before then evaluating each unit on a specific aspect. I demonstrate the framework's effectiveness at evaluating factuality and faithfulness on tasks such as book summarization and biography generation. Overall, our experiments suggest that despite their impressive capabilities, LLMs have a long way to go before their output can be trusted.