Researchers at Mila have proposed a new technique that makes large language models (LLMs) vastly more efficient when performing complex reasoning. Called Markovian Thinking, the approach allows LLMs ...