Research shows that when LLMs are asked to think step-by-step before providing answers (Chain-of-Thought prompting), their performance on complex reasoning tasks improves dramatically.