Rahid Zahid Alekberli

The paper introduces the Kerimov-Alekberli model, an information-geometric framework that uses non-equilibrium thermodynamics and stochastic control to provide a physically grounded method for detecting AI safety anomalies in real-time.

The paper proposes a novel information-geometric framework to analyze LLM stability by integrating task utility, external entropy, and internal structural proxies, showing this composite score improves reliability assessment over simple baselines.