Tag
This paper proposes a method to detect hallucinations in LLMs by analyzing topological signatures in attention graphs, showing improvements over existing baselines across multiple benchmarks.