Computer Science
Hallucination Detection in LLMs Using Spectral Features of Attention Maps
J. Binkowski, D. Janiak, et al.
Discover LapEigvals — a new spectral approach that treats attention maps as graph adjacency matrices and uses top-k Laplacian eigenvalues to detect LLM hallucinations. Experiments show state-of-the-art performance among attention-based methods, with strong robustness and generalization. This research was conducted by the authors present in the <Authors> tag.
Related Publications
Explore these studies to deepen your understanding of the subject.
