Formalizes the 'Neural Uncertainty Principle,' linking adversarial vulnerability in vision and hallucinations in LLMs to a shared geometric and information-theoretic origin.
March 23, 2026
Original Paper
Neural Uncertainty Principle: A Unified View of Adversarial Fragility and LLM Hallucination
arXiv · 2603.19562
The Takeaway
By viewing these two major AI failures as two sides of the same 'uncertainty budget' coin, the paper provides a unified diagnostic tool. It offers practical, training-free methods (like ConjMask and prefill-stage probes) to detect hallucination and improve robustness.
From the abstract
Adversarial vulnerability in vision and hallucination in large language models are conventionally viewed as separate problems, each addressed with modality-specific patches. This study first reveals that they share a common geometric origin: the input and its loss gradient are conjugate observables subject to an irreducible uncertainty bound. Formalizing a Neural Uncertainty Principle (NUP) under a loss-induced state, we find that in near-bound regimes, further compression must be accompanied by