Advanced Prompting Techniques

Introduction

Once you master the basics, you can use advanced techniques to tackle complex reasoning tasks and dramatically reduce hallucinations.

Few-Shot Prompting

Instead of just asking the model to do something (zero-shot), you provide a few examples of the input-output pairs you expect. This "trains" the model in-context.

Extract the names of cities from the text.
Text: I visited Paris last summer.
Cities: Paris

Text: She moved from Tokyo to New York.
Cities: Tokyo, New York

Text: We are flying to Berlin tomorrow.
Cities:

Chain of Thought (CoT)

Adding a phrase like "Let's think step by step" forces the LLM to output its reasoning process before giving the final answer. This drastically improves performance on math and logic puzzles.

Assignment