🟢 Introduction to Few-Shot Prompting Techniques

Welcome to the Few-Shot section of the advanced Prompt Engineering Guide.

While Zero-Shot prompting is the most basic form of interaction—where the Large Language Model (LLM) Few-Shot prompting techniques it a step further. Few-Shot prompting provides the model with example pairs of problems and their correct solutions. These examples help the model better understand the context and improve its response generation.

Here are a couple of techniques we've already explored, with more on the way!

• K-Nearest Neighbor (KNN) Prompting selects relevant examples by finding the most similar cases to the input query, improving the accuracy of Few-Shot prompts.
• Self-Ask Prompting breaks down complex questions into sub-questions, and helps LLMs reason more effectively and provide better answers.
• Prompt Mining selects the optimal prompt template for a given task from a corpus of text based on the template that comes up most often in the corpus.
• Vote-K Prompting selects diverse and representative exemplars from unlabeled datasets for Few-Shot prompts.

Stay tuned for more advanced techniques, including:

• SG-ICL

🟢 Chain-of-Dictionary (CoD)

🟦 Chain of Knowledge (CoK)

🟢 Cue-CoT

◆ K-Nearest Neighbor (KNN)

◆◆ Prompt Mining

🟢 Self-Ask

🟢 Self Generated In-Context Learning (SG-ICL)

◆◆ Vote-K