Language models often stumble when a question needs several steps of thinking, such as a math problem, a logic puzzle, or a decision with trade-offs. Reasoning techniques are ways of prompting that give the model room to work through a problem instead of jumping straight to an answer.
You'll start with chain-of-thought prompting, where the model writes out its steps, and its zero-shot version, which asks the model to think step by step without any examples. Then you'll learn self-consistency, which compares several reasoning paths to find the most reliable answer, and finish with prompt chaining, which splits a big task into a series of smaller prompts.
Once a model can reason well, you'll want its answers in a form your code can use. That's the focus of Structured Outputs & Formatting, the next module in this course.
What chain-of-thought prompting is and how it can be used to improve prompts to large language models.
How zero-shot chain-of-thought makes a language model write out its reasoning step by step before giving a final answer.
Everything you need to know about self-consistency prompting, with the latest research, templates, and examples.
Why one giant prompt can make the output fall apart, and how prompt chaining breaks a task into focused steps.