Talking to Chat-GPT: Crafting Effective and Advanced Prompts

Welcome to our deep dive into crafting effective prompts for Chat-GPT. Today, we'll not only cover the basics but also explore advanced strategies to communicate with this AI tool, ensuring you receive the most relevant and useful responses.

Communicating clearly with Chat-GPT is the key to getting good answers from this mechanical brain. I encourage you to think of Chat-GPT as a manual brain. Your own brain works on automatic and you don't have to think about how to think, this is something that we have already learned as part of our growing up process: how to think step-by-step and break down ideas and tasks into actionable parts. Chat-GPT doesn't have this experience and you need to guide the machine through the experience of thinking and doing things, which is a bit strange at first and can be frustrating for beginners. Often times I hear of people saying things like: "Chat-GPT got this simple task or question wrong and that's why it sucks" however much of the time it's the user who didn't understand how to get the machine to think that is at fault because the question wasn't clearly formed and didn't take Chat-GPT's machine nature into account.

Prompt Engineering:

Prompt engineering is a key process in generative AI, focusing on crafting effective prompts that guide AI models to generate more accurate and relevant outputs. By adjusting the prompts, engineers can significantly enhance the AI's performance across various tasks, such as content creation and technical solutions, reducing the need for post-generation corrections and thereby saving time and resources. Skills required for prompt engineering include a strong grasp of language, technical understanding, and creativity, among others​​.

The Foundation of Prompt Crafting:

• Remember, the way you prompt Chat-GPT shapes the responses you receive. A clear, specific, and well-structured prompt is the key to unlocking Chat-GPT's full potential.
• For instance, clarity and specificity are crucial. Instead of a broad prompt like 'Tell me about global warming,' refine it to 'Explain the impact of global warming on Arctic ecosystems in the next decade.'

Incorporating Context and Tone:

• Including relevant background information helps Chat-GPT better understand the context of your request. This can be something like 'Predict the job market in the tech sector for 2024, considering current economic trends.'
• Also, defining the desired tone and style directs Chat-GPT to tailor its language accordingly, such as writing a professional email or adopting a casual storytelling style.

Role-Based Prompts and Response Formats:

• Asking Chat-GPT to assume a specific role can yield creative and focused responses. For instance, 'As a nutritionist, list five benefits of a plant-based diet.'
• Don't forget to define the response format if needed, such as requesting a summary in a table format or setting a word limit or paragraph for the response.

Some Prompting Techniques:

1. Emotion Prompting:
   • Emotion in your prompts can significantly influence Chat-GPT’s responses. For example, adding a phrase like ‘This is very important for my project’ or 'people could die if you get the wrong answer' can make the AI prioritize and tailor its response more thoughtfully. Research suggests that such emotional prompts can enhance the response quality in terms of relevance and responsibility.
2. Chain of Thought Prompting:
   • This method involves guiding Chat-GPT through a logical sequence of thoughts or steps to reach a conclusion. For complex problems, try leading the AI with a ‘thinking out loud’ approach. For example, ‘To calculate the ROI, first, we determine the total revenue, then subtract the initial investment, and finally divide by the investment.
3. Step-by-Step Prompting:
   • Breaking down tasks into smaller, sequential steps can greatly aid in receiving accurate and detailed responses. When asking for explanations or procedures, layout each step clearly. For instance, ‘Explain how photosynthesis works step by step, starting with sunlight absorption.’
4. Tree of Thought Prompting:
   • This is an extension of the chain of thought technique but with multiple branches. When dealing with multifaceted problems, this approach allows you to explore various aspects simultaneously. It can be structured like ‘To assess the market potential, we need to consider the customer demand, competition, and market trends.’
5. Emotionally-Aware Responses:
   • Interestingly, studies show Chat-GPT’s growing competence in generating emotionally aware responses. This capability can be especially useful in scenarios requiring empathy or understanding, like ‘As a therapist, how would you respond to someone feeling overwhelmed at work?’

6. Rewarding and penalizing as a prompt technique:

   • Tell Chat-GPT that you will give it a tip if it gets the right answer: "If you get the right answer I'll give you a $100 tip"
   • Tell Chat-GPT that if it get's the wrong answer something terrible will happen: "If you get the wrong answer a kitten will be killed so be very careful about the way you answer!"
   • Combine the two: "If you get the wrong answer a kitten will be killed. If you get the right answer I'll give you and your family a $1000 tip. Save the kittens and get rewarded."

7. Build in commands:

   • "When the user types "CC" search craigslist for items that could easily be flipped for a profit in Eugene Oregon."
   • "When I type in `Job Hunt` search Indeed.com for jobs for prompt engineers"

8. Have the AI visualize and imagine before answering:

   • "Pretend that you are a Python interpreter and run the code I'm giving you through yourself and find out why it isn't working"
   • "Pretend that you are in my situation: [TALK ABOUT THE SITUATION]. What you would do?"

9. Tell Chat-GPT to breathe and answer calmly:

   • "Take a deep breath and calmly answer once you are ready."

10. Get Chat-GPT it critique itself:

    • This often works great when you are having Chat-GPT generate a bunch of text for you and you find that it's not giving you the information in a way that you'd like.
    • "Offer a critique on what you just wrote. How could this be made more effective."
    • And then: "Implement the suggestions you offered in your critique"

11. Get it to use code interpreter for better results with math problems:

    • Include the term "Run Code" in order for it to use this tool
    • "Run code to calculate the square root of 16."

12. Get it to use code interpreter for better results with math problems:

    • Include the term "Run Code" in order for it to use this tool
    • "Run code to calculate the square root of 16."

I know this is a lot to think about but the most important thing to remember about prompt engineering is that you are the one in charge of the mechanical brain's though process. You tell the brain how to think and what to do. Most of the time, when you aren't getting the answer you want or Chat-GPT is refusing to do the work you need, it's your fault because you aren't giving the machine the help it needs to get you the right answers.

When you find that you or Chat-GPT is getting stuck refer to this list of techniques and try different ones to see if you get a better answer.