LLMs simulate reasoning step-by-step, but their logic is predictive, not truly sequential. New models use “continuous thought ...