FLaG: Fine-Grained Latent Grouping for Hallucination Detection

In three linesFLaG is a lightweight hallucination detection framework for LLMs that models correctness through latent evidence groups. Using energy-based routing and log-marginal aggregation, it captures heterogeneous hallucination patterns without modifying the underlying model. SOTA results across multiple benchmarks with robust transfer across datasets.

Summary generated by Claude — human-verified