PROMPT TUESDAY: One Simple Trick to Get Better ChatGPT Results
And Why AI Often Stinks at Its Responses
Yes, ChatGPT often stinks.
(Feel free to use Claude, Gemini, Mistral, Llama, etc., interchangeably with ChatGPT for the rest of this piece.)
It’s often impatient.
It gets in such a rush to deliver any result that it skips over interesting topics or goes too far down rabbit holes on points you care nothing about.
Fortunately, you can use its speed and understanding to your advantage.
When I write a blog post, an answer to a Quora question, a LinkedIn update, or a book, I don’t have to wait forever for it to create new drafts or tweak problems in outputs that I otherwise enjoy.
I can zero in on the parts that I don’t like or the parts that need further thought.
For this inaugural edition of PROMPT TUESDAY—yeah, I know, I need a better title—I’ll demonstrate with a Quora response I submitted today.
The question: Will we ever be able to train an AI on human nature?
To start, I had some thoughts.
What is human nature?
Human nature in China looks a heck of a lot different than it does in the US.
One nation emphasizes contributions to the whole; the other focuses on individuality.
Furthermore, “Human Nature” means different things depending on the stage of life that you’re in. My Nature at 45 was nothing at all like my Nature at 25.
I’m willing to bet the same is true for all of you lucky enough to live this long.
And, to get a real understanding of Human Nature, we would need data sets to analyze. Unfortunately, social and psychological sciences aren’t comprehensive or infallible. They’re also filled with biases.
I know this, but I’m not an expert on it. That’s where I needed some help from my AI friend, ChatGPT. Here’s what it initially came up with:
Much of what we think we know about human behavior is shaped by limited studies, often focused on specific populations. If we train an AI on those biases, we’d risk creating something error-prone and harmful when applied on a broader scale.
Not bad, but there’s one little problem, ChatGPT. You fail to give me an example of what the heck you’re talking about.
If I post that to Quora, people will see right through it. So, enter a simple follow-up question:
Give an example of this: [FOLLOWED BY A PAST OF THE ABOVE PARAGRAPH HERE]
ChatGPT took only a few seconds to deliver the goods. It came up with this:
Here’s a concrete example:
Consider psychological research on intelligence, particularly the early studies that shaped concepts like IQ testing. Many of these studies were based on populations in Western, industrialized societies—commonly abbreviated as WEIRD (Western, Educated, Industrialized, Rich, and Democratic). These populations represent a small fraction of human diversity, yet findings from such studies have often been generalized to humanity as a whole.
If we trained an AI to predict or evaluate intelligence based on this data, it could embed the biases inherent in the testing methods. For instance:
Cultural Bias: Standardized IQ tests often rely on language, logic, and problem-solving patterns familiar to WEIRD populations but may disadvantage people from other cultural or socioeconomic backgrounds.
Historical Bias: Many early studies reflected the prejudices of their time, linking intelligence to racial or ethnic groups in ways we now recognize as deeply flawed.
Training an AI on such biased data could lead to harmful outcomes. Imagine deploying this AI in hiring systems, academic admissions, or policy decisions. It might systematically undervalue intelligence or potential in non-WEIRD populations, perpetuating discrimination and inequality.
This isn’t just a hypothetical concern. AI systems in other fields, like facial recognition, have already shown how bias in the training data—often favoring lighter skin tones—can result in poor performance or outright harm to underrepresented groups. It’s a cautionary tale for why training AI on incomplete or skewed datasets can be both error-prone and socially damaging.
Now, I was quite happy with most of my Quora Answer, so I didn’t use most of this. But, the detailed example that it provided about the biases in IQ tests was enough to get me over the creative hump.
I used enough of it to extend my thoughts and round out the Answer with a good example of what I was talking about.
So, to fulfill the promise of this post, the ONE SIMPLE THING you can do to get better results out of your AI outputs is to ask for examples.
This forces the AI to think harder about interesting topics it may have glossed over in the first output.
Beyond that, here are some action steps to help you maximize your AI tool’s potential when crafting prompts and refining outputs:
1. Start with Clear, Context-Rich Prompts
Instead of vague commands like “Explain human nature,” provide specific context and desired outcomes: “Explain how human nature differs across cultures and life stages, focusing on individualism vs. collectivism.”
The clearer your prompt, the less time you’ll spend revising AI’s initial output.
2. Break It Down into Smaller Tasks
If the result feels rushed or superficial, break your query into parts. For example:
“Define human nature.”
“List cultural differences in how human nature is expressed.”
“How does human nature change with age?”
This keeps the output focused and reduces the chances of tangential rabbit holes.
3. Ask for Examples
AI often speaks in generalities. Prompt it to provide concrete examples. Use simple follow-ups like:
“Can you give me a specific example of what you mean?”
“Illustrate this with a historical or cultural case.”
This approach enriches the response and makes your work more relatable and impactful.
4. Iterate and Layer
If an output feels incomplete, dig deeper:
“Expand on how these biases in IQ tests might influence modern AI applications in hiring or education.”
“What solutions have researchers proposed for addressing such biases in AI training?”
Use AI as a brainstorming partner that builds on each layer of your inquiry.
5. Customize the Style and Depth
Tailor the tone or detail level of the response:
“Summarize this for a high school audience.”
“Make this answer sound more conversational.”
“Provide a technical analysis suitable for a white paper.”
6. Use AI for Polishing, Not Perfection
Rarely will the first response be exactly what you want. Zero in on weak spots and prompt fixes:
“Rewrite the section on IQ test biases to focus on cultural diversity.”
“Simplify this paragraph for clarity without losing detail.”
7. Experiment with “Priming”
Before diving into a topic, warm up the AI:
“Imagine you’re an expert sociologist explaining human nature to a curious audience.”
This primes the AI to focus on relevant content and adopt the right tone.
8. Pull from Multiple AI Tools
If one tool misses the mark, try another.
Claude and Gemini are great for conversational or contextual nuance, while Mistral and Llama excel in detailed, technical outputs.
You can combine their strengths by comparing responses and integrating the best parts.
9. Stay Open to Surprises
AI’s value isn’t just in doing what you ask—it’s also in offering perspectives you might not consider.
Explore those rabbit holes; they might yield insights you didn’t realize you needed.
10. Practice Makes Perfect
The more you engage with AI tools, the better you’ll get at crafting prompts that generate strong results.
Review old outputs, refine your approach, and experiment with different phrasing.
The more intentional and interactive you are with your AI tool, the better it will perform.
Instead of waiting for the perfect draft, think of AI as a partner that helps you refine, expand, and deepen your ideas quickly and efficiently.
MORE TO COME!
You’ve received two posts in a row from me this week. There’s a reason for that.
I’m trying to make this channel worth your while. So, every Monday-Friday, we’ll focus on something different.
Mondays will be news-driven.
Tuesdays will focus on the craft of prompting.
On Wednesday, we’ll spend more time with AI tools—what they can do, how to use them, etc.
Thursdays, we’ll cut loose with some cool AI creations in the world of audio, video, and image creation.
And, on Fridays, we’ll bring it home with how you can implement AI in your specific niche or industry.
This is in addition to the quick news and tools updates I’ve got going at Innovation Dispatch (also M-F, SUBSCRIBE HERE), and my just-launched LinkedIn newsletter, mAIn Street (SUBSCRIBE HERE).
ID gives you just enough AI news and tools to stay on top of where the industry is. Substack will take you deeper into how to use and apply AI.
Last but not least, mAIn Street will give you weekly updates on how AI and modern technology will shape our world in various industries.
Soon, I’ll also be adding some premium content.
I intend to monetize this channel soon, but all this new content will ensure you’re still getting value even if you choose the Free tier only.
So, that’s the plan. I hope you will subscribe to one, two, or all three. I love connecting with you! Until tomorrow…