Researchers at Carnegie Mellon University have developed a novel training technique for large language models (LLMs), allowing developers to control the chain-of-thought length for more…