An LLM can only call out a premise if the average response to an input in its model calls out the input.