How can an LLM switch between low-, medium-, and high-effort reasoning? And how does an LLM learn to reason more or less? I put together a “little” article explaining how these effort levels are implemented at inference time and during training.