The Most Typical Errors People Make When Using AI
Two reminders my brother gave me by accident
For the first time in a very long time, my entire family is in my childhood home. My brother’s unsure of where and when he’ll leave to some other country. I haven’t found a flat yet.
So we’ve been working in different corners of the house and often end up chatting about work. My brother’s in finance. He creates crazy formulas to fill in enormous Excel files with yet more numbers. He’s also quite tech-savvy through pure curiosity and tons of trial and errors.1
I’m also quite tech-savvy but in a different way. I’ve played with many more tools than him and plunged into AI earlier and deeper than he ever did. My tests and research for this newsletter have only deepen further my knowledge and using GPT-4 gives me an “advantage” when looking to bypass a hurdle.
For the past year or so, my brother’s gotten used to turn to AI to help him create more monstruous formulas and automatizing more his Excel files but one was really bugging him the other day so—out of desperation2—he asked me if we could check with GPT-4.
So we did. And so we gave up after results weren’t all that better.3
Still. I don’t like a mystery unsolved so I kept going on my own for a little while longer. One of the results I got was a formula that, well, didn’t work any better than the rest.
But it felt different.
I looked back at what he had been using and figured a combination of the two formulas might work. And it did! 🥳
In about 20 minutes of research and thinking, I had unblocked my brother’s problem on which he had spent the previous 5 hours. He was happy with it and moved forward.
Now, let’s be clear. Without his original formula and explanation of its components, I’d have never figured this out. But the big difference was our approach to Chat-GPT.
He looked for a solution. I looked for a hint to the solution.4
A week passed and another formula arrived.
This time, he knew precisely what he wanted to do but had no potentially working formula despite having spent a full day on it. We tried some time and all the proposed formulas used a formula called LAMBDA
. He said that wasn’t a real formula but one we had to set up so it wouldn’t work. He gave up and left the room.
I kept looking. I pushed Chat-GPT again and again to give me a formula without LAMBDA
nor without using VBA but it kept going back to it despite my directives.
After a while, I actually looked into LAMBDA
and how it worked. It turned out, it could work and wasn’t all that complicated either.
I called my brother again and explained my thoughts. He was dubitative but it did the job so he chose to use it in the end.
That was yet another reminder that sometimes, what we think won’t work, well, can.
TL;DR
The most common error people do when using AI is to want the solution right away. One prompt = one answer. Maybe they’ll keep a conversation going for 4 or 5 exchanges but they rarely go further, even though pushing and getting more answers that aren’t what you want might actually be useful.
AI chatbots aren’t perfect. Far from it. They’re a good support.
A good place to start.
Don’t forget you have your own knowledge and reflection at your disposal.
Use what you already know to see how what you received could be tweaked to get your goal. And if you keep pushing and keep getting a similar response, it might just be because that’s the way to go—whether you want it or not.
Does this apply to language learning?
Now. Languages are a lot more flexible than Excel formulas. There rarely—if ever—is a single true answer.
If you’re learning a language, however, you already have the knowledge from the past lessons and exposure you’ve gone through. You’ve combined words and patterns in multiple ways. What you need might just be another way to combine some of them to notice how your knowledge can be useful.
And sometimes, a pattern you’re trying to avoid might just be the only one that doesn’t require a complete restructuration of the sentence.
I love to ask Chat-GPT for other ways to say something I know how to say because it helps me see how patterns I think I’ve mastered can be used in more creative ways.
So, yeah, use AI all you want to learn.
But don’t forget to use your head along the way.
Cheers for reading,
Mathias
In his first job, about 15 years ago, he learned (on his own) how to create macros in Excel and built a 78-page macro that did helped him finish his daily workload in about 5 min.
After trying GTP-3.5, Claude, Gemini, tried tons of modifications, and gotten pretty angry at the lack of result
GPT-4’s quality seems to have dropped by quite a margin these past few months, making me question my subscription.
Well, that’s what I did after realizing I wouldn’t get the actual solution from it.