FACTORY: A Challenging Human-Verified Prompt Set for Long-Form Factuality
This addresses the problem of unreliable benchmarks for evaluating long-form factuality in language models, though it is incremental as it builds on existing evaluation methods.
The authors tackled the lack of human verification in long-form factuality evaluation by introducing FACTORY, a large-scale, human-verified prompt set, and found that approximately 40% of claims from state-of-the-art models were not factual, compared to 10% for other datasets.
Long-form factuality evaluation assesses the ability of models to generate accurate, comprehensive responses to short prompts. Existing benchmarks often lack human verification, leading to potential quality issues. To address this limitation, we introduce FACTORY, a large-scale, human-verified prompt set. Developed using a model-in-the-loop approach and refined by humans, FACTORY includes challenging prompts that are fact-seeking, answerable, and unambiguous. We conduct human evaluations on 6 state-of-the-art language models using FACTORY and existing datasets. Our results show that FACTORY is a challenging benchmark: approximately 40% of the claims made in the responses of SOTA models are not factual, compared to only 10% for other datasets. Our analysis identifies the strengths of FACTORY over prior benchmarks, emphasizing its reliability and the necessity for models to reason across long-tailed facts.