In my last article AI, Consciousness, and the Future of Intelligence, we looked at some interesting philosophical ideas around AI. (If you’re a fan of Blade Runner or Westworld, check it out)
In this one we’ll look at a practical use case for making your AI tools more “aware”. Specifically a technique that can significantly enhance your interactions with ChatGPT.
Chain of Thought Prompting
Simply put, Chain of Thought Prompting is a way to ask better questions (and get better results) with AI.
Chain of Thought Prompting isn’t just a fancy term; it’s a systematic approach to getting more nuanced, coherent, and accurate responses from generative AI models.
Think of it as guiding the AI through a topic or question, ensuring it understands what you’re trying to do and helping it process information.
Why Use Chain of Thought Prompting?
- For Complex Topics:
Cultural differences can be broad and varied with lots of room for nuance. Using Chain of Thought Prompting can help get a well-rounded view instead of a generalized answer. If you want to use AI to become more cultured and avoid accidentally offending someone, it’s worth a try.
2. For Better Results:
A more specific question will lead to a more specific answer, clarifying follow up questions can do even more. Useful information and the chance to reflect on previous answers can work wonders (for chat bots and people). This technique helps us guide the AI to the details we really want and put them in the right format.
Context Windows in Chain of Thought Prompting
If you use ChatGPT often, this is something you must know.
Every large language model, including ChatGPT, operates within a “context window”, essentially its working memory. Think of it as the AI’s short-term recall.
ChatGPT has a context window of about 4,000 words. This means that any parts of a conversation that is past this point is effectively “forgotten”. This can make it tricky to do more complex tasks, but for most use cases, 4000 words is plenty.
Side Note: If you’d like to try a newer chatbot with a bigger context window, check out Claude 2. It’s an AI tool similar to chatGPT but it’s trained with more up to date information and has a context window of roughly 75,000 words!
When you use Chain of Thought Prompting and continue within the same conversation thread, the AI can reference previous prompts and messages, leading to more coherent and contextual answers.
For example, after discussing dining etiquettes in the USA, if you immediately ask about those in Japan, the AI can contrast them more effectively, having the prior context fresh in its memory.
Key Takeaway: Stick to the same conversation thread. It’s like having a focused chat with a human — the flow remains uninterrupted, and the context is retained.
How to Use Chain of Thought Prompting
1. Break Down Your Question:
Instead of asking, “Tell me about cultural differences between the USA and Japan”, try:
- “What are common greetings in Japan compared to the USA?”
- “How do work ethics differ between the USA and Japan?”
- “What are dining etiquettes in both countries?”
Ask these questions one by one, in the same chat. By doing this, you can expect more detailed responses.
2. Be Specific, Include Context:
ChatGPT can talk about anything, if you’d like it to do something specific, be specific. Add any context that it might find useful, long prompts with more detail are usually better.
3. Give feedback! :
Remember, ChatGPT is not a google search, it’s a chat! Keep the conversation going. If it’s given you an answer and you want something different, just say so.
Let’s try an example; You want to send a short email to a friend and colleague before a business trip to Japan.
Not bad, but it’s pretty long, it’s also a bit too formal and missing info. Let’s give some feedback and additional details about the meeting:
Better! But now it’s a a little too casual, and I’d prefer to start with the pitch deck. Plus, a friend told me about a sushi restaurant in Tokyo she really enjoyed. Let’s keep going:
Much better! Instead of taking the first answer as a given, we worked with GPT and gave it the chance to improve based on our feedback.
Different Ways to Use Chain of Thought Prompting
1. Hierarchical Prompting:
Start general, then dive into specifics. For instance:
- “What are cultural norms?”
- “How are cultural norms in India different from France?”
- “What gestures are considered rude in France but okay in India?”
2. Comparative Prompting:
Have the AI contrast things. Example:
- “How is family structure in Mexico different from Canada?”
3. Thought Process Prompting:
Ask it to write out it’s thought process, try adding these to the end of your prompts:
- “…, think step by step”
- “…, thinking aloud”
- “…, explain your thoughts”
4. Sequential Prompting:
Ask for a sequence or progression.
- “How have British cultural norms evolved over the last century?”
Benefits of Chain of Thought Prompting
Using this method with ChatGPT helps in obtaining clearer and more comprehensive answers. It shines especially when we tackle vast subjects like cultural differences, ensuring we get a diverse range of information.
If you want to use AI tools effectively, you need the right approach. Chain of Thought Prompting offers a way to make our interactions with tools like ChatGPT more fruitful. As our reliance on AI grows, techniques like these become crucial to understand and practice.
Tried using Chain of Thought Prompting with cultural topics on ChatGPT? Share your thoughts below. Or suggest another topic you’re curious about. There’s always more to explore with AI.