March 20, 2025
I asked ChatGPT’s “Deep Research” how many people know about Pi Day.
Silly? Yes. But more useful than you might think.
In this article, you will learn:
- What Fermi problems are and why they matter in interviews
- How AI tools approach estimation challenges
- The surprising depth of analysis possible with simple prompts
- Practical applications for AI-assisted research
- How these tools can enhance your own problem-solving skills
Interview Challenge
In one of my first interviews, I was asked, “At this exact moment, how many golf balls are in the air in South Africa?” I had to think out loud and arrive at an actual number.
I’ve since enjoyed asking similar questions from the other side of the table. You might have heard examples like:
- Liters of paint needed to repaint the Eiffel Tower
- Gas stations in New York
- Pizzas ordered in Cape Town
These are “Fermi Problems,” designed to test how well you break down complex problems using logical approximations.
Fermi Problems
Answering well can teach the interviewer a lot about you.
It shows logical reasoning, estimation skills, assumption setting, problem-solving, ability for back-of-the-envelope calculations (in your head!), creativity, and lateral thinking. And, importantly, confidence.
Some people smashed it. But we’ll save great interviews for another day.
AI to the Test
Now, to celebrate Pi Day, I gave ChatGPT a similar challenge. I asked: “What proportion of people in the world know about Pi Day?” and to share its reasoning, assumptions, and calculations.
What my $20-a-month research assistant did in a few minutes on such a trivial matter was eye-opening.
It has one huge advantage over my poor interviewees: access to the internet. But you can learn so much from the way it approaches the problem, breaks it up into chunks, considers elements that might have an impact, makes assumptions where it needs to, and does the calculations.
Depth of Analysis
To give you just a few examples of what it considered and reported on from an extremely simple prompt:
- US surveys on Pi Day and Pi itself (The things they keep themselves busy with!)
- Date format differences in countries (3/14 vs. 14/3)
- The levels of math and science education in different countries
- STEM career interest and its impact
- Internet reach and global differences (Apparently, Pi Day is a hit on TikTok?)
- Google Trends and social media hashtags
- Language and cultural contexts
It calculated Pi Day awareness in different continents, ultimately estimating 4% globally and 5% in South Africa.
Practical Applications
Would I take this straight to a boardroom? Probably not.
But in just a few minutes, I could:
- Supplement my own thought process
- Consider assumptions I hadn’t thought of
- Get useful data points and sources that would’ve taken me ages to find
Future of Research
Research has already changed for all of us. If you’re not experimenting with these tools yet, start today.
It might even help you with your next Fermi Problem.