A single-pass black-box method models LLM outputs as dynamical systems via Koopman operators to detect hallucinations with claimed state-of-the-art accuracy and lower cost.
Title resolution pending
2 Pith papers cite this work. Polarity classification is still indexing.
verdicts
UNVERDICTED 2representative citing papers
The work defines slow manifolds via unstable-stable manifold intersections and applies the isostable framework to produce reduced models for Hodgkin-Huxley neuron networks and a detailed circadian oscillator, illustrated on two control tasks.
citing papers explorer
-
Low-Cost Black-Box Detection of LLM Hallucinations via Dynamical System Prediction
A single-pass black-box method models LLM outputs as dynamical systems via Koopman operators to detect hallucinations with claimed state-of-the-art accuracy and lower cost.
-
Reduced Order Modeling of Nonlinear Dynamical Systems Using Slow Manifolds
The work defines slow manifolds via unstable-stable manifold intersections and applies the isostable framework to produce reduced models for Hodgkin-Huxley neuron networks and a detailed circadian oscillator, illustrated on two control tasks.