“Martin Casado believes that AI models are not effective for predictive tasks, as they are fundamentally designed to provide a stochastic smoothing of their past training data.”