Skip to main content

Lily's Lab

Back to all posts

Understanding Key Prompting Techniques

Artificial-Intelligence

Welcome to prompt engineering! It's all about crafting the right text inputs, or prompts, to get the best responses from language models (LLMs). Think of it as guiding a super-smart AI to understand what you need. Here are some key techniques to get you started:

→ General Prompting / Zero-Shot: This is the simplest way. You give the LLM a task description without any examples. It's like asking a question directly.

  • Example: "Classify this movie review as POSITIVE, NEUTRAL, or NEGATIVE. Review: 'This film was absolutely amazing!'". The LLM should ideally output "POSITIVE".

→ One-Shot & Few-Shot: When zero-shot isn't enough, you can provide examples in your prompt.

  • One-Shot: You give one example to show the LLM what you're looking for.
    • Example: "Translate 'Hello' to French. Bonjour. Translate 'Thank you' to French." The LLM should output "Merci".
  • Few-Shot: You provide multiple examples (usually 3-5) to show a pattern. This helps the LLM understand the task better.

    • Example:
      The LLM should complete with "Tokyo".

      Question: What is the capital of France? Answer: Paris.
Question: What is the capital of Germany? Answer: Berlin.
Question: What is the capital of Japan? Answer:

→ System, Contextual, and Role Prompting: These techniques help you guide the LLM in different ways.

  • System Prompting: You set the overall context or goal for the LLM. It defines the 'big picture'.
    • Example: "Classify movie reviews as positive, neutral or negative. Only return the label in uppercase. Review: 'It was a terrible movie.' Sentiment:". The LLM should output "NEGATIVE".
  • Contextual Prompting: You provide specific background information relevant to the current task. This helps the LLM understand the details.
    • Example: "Context: You are writing for a blog about retro 80's arcade video games. Suggest 3 topics to write an article about...".
  • Role Prompting: You assign a specific persona or role to the LLM. This influences the style and voice of its responses.
    • Example: "I want you to act as a travel guide... My suggestion: 'I am in Amsterdam and I want to visit only museums.' Travel Suggestions:".

→ Step-Back Prompting: You first ask the LLM a general question related to the specific task to activate background knowledge, then use that answer in the main prompt.

  • Example:
    • Step-Back Prompt: "Based on popular first-person shooter action games, what are 5 fictional key settings that contribute to a challenging and engaging level storyline...?".
    • (LLM's Response - excerpt): "1. Abandoned Military Base...".
    • Final Prompt: "Context: 5 engaging themes for a first person shooter video game: 1. Abandoned Military Base... Take one of the themes and write a one paragraph storyline...".

→ Chain of Thought (CoT): You prompt the LLM to explain its reasoning step by step before giving the final answer. This helps it solve complex problems.

  • Example: "When I was 3 years old, my partner was 3 times my age. Now, I am 20 years old. How old is my partner? Let's think step by step.".

→ Self-Consistency: You generate multiple reasoning paths (using CoT with a higher randomness) and choose the most common answer. This improves accuracy.

→ Tree of Thoughts (ToT): This advanced technique allows the LLM to explore multiple reasoning paths simultaneously.

→ ReAct (reason & act): The LLM combines reasoning with the ability to take external actions (like using search engines) in a loop to solve tasks.

→ Automatic Prompt Engineering (APE): You use an LLM to automatically generate and evaluate different prompt variations.

![[LilysLab-Logo.png]]