Tag
This paper argues for a sequential inference framework to enhance LLM trustworthiness by modeling interactions as dependent stochastic processes, ensuring validity under repeated use, and enabling online monitoring for behavioral shifts.