Paper · arXiv 2408.07852
Training Language Models on the Knowledge Graph: Insights on Hallucinations and Their Detectability
Jiri Hron, Laura Culp, Gamaleldin Elsayed, Rosanne Liu, Ben Adlam, Maxwell Bileschi, Bernd Bohnet, JD Co-Reyes, Noah Fiedel, C. Daniel Freeman, Izzeddin Gur, Kathleen Kenealy, Jaehoon Lee, Peter J. Liu, Gaurav Mishra, Igor Mordatch, Azade Nova, Roman Novak, Aaron Parisi, Jeffrey Pennington, Alex Rizkowsky, Isabelle Simpson, Hanie Sedghi, Jascha Sohl-dickstein, Kevin Swersky, Sharad Vikram, Tris Warkentin, Lechao Xiao, Kelvin Xu, Jasper Snoek, Simon Kornblith
Larger language models trained on fixed datasets exhibit fewer hallucinations, but detecting these hallucinations becomes more difficult as model size increases, despite improvements in detector size.
Abstract
While many capabilities of language models (LMs) improve with increased training budget, the influence of scale on hallucinations is not yet fully understood. Hallucinations come in many forms, and there is no universally accepted definition. We thus focus on studying only those hallucinations where a correct answer appears verbatim in the training set. To fully control the training data content, we construct a knowledge graph (KG)-based dataset, and use it to train a set of increasingly large LMs. We find that for a fixed dataset, larger and longer-trained LMs hallucinate less. However, hallucinating on leq5% of the training data requires an order of magnitude larger model, and thus an order of magnitude more compute, than Hoffmann et al. (2022) reported was optimal. Given this costliness, we study how hallucination detectors depend on scale. While we see detector size improves performance on fixed LM's outputs, we find an inverse relationship between the scale of the LM and the detectability of its hallucinations.