I have a gut feeling the particular stochastic processes we’re currently using are the wrong ones to go beyond LLMs into actual intelligence.