Data, data, and more data.
That's what we've been told for over a decade. That we need data in our businesses.
And for what purpose?
Well, to make decisions based on it. Informed decisions. Decisions that have a strong foundation. Decisions that take into account the past, present, and future of the company.
But all those decisions are nothing without a main component: mathematics.
Mathematics helps us have a common language, to abstract concepts, to project them into the future, to establish steps to make the business more efficient.
And with all the hype around AI, you might wonder why you need mathematics.
Today on Feasible, I'll tell you:
The challenges AI faces with mathematics
The power of solvers when it comes to solving complex decision problems
How AI and solvers can be integrated to optimize decision-making in your business
So yes… Let's go for it!
🤖 Understanding AI’s Mathematical Challenges
It's amazing how well the responses generated by ChatGPT, Claude, Gemini (ahem, ahem)...
They seem to understand perfectly what we ask them. We read, we ask things et voilà. It seems like magic.
But that magic vanishes when you encounter things like:

You might tell me I'm cherry-picking. It's true that ChatGPT now tells you that 9+10 is 19. But what happens if we make it do more complex multiplications?
This is something I'm not the only one wondering about.
A few months ago, researchers from Tsinghua University (Beijing) - one of the world's most prestigious universities, within the Top 20 - decided to ask the same question.
They concluded that the technology behind ChatGPT is capable of performing mathematical operations, although looking at the results, I have my doubts.
Let me explain.
The best results are obtained with operations that include 5-digit numbers, and their accuracy is about 85%. That is, they fail in 15 out of every 100 cases. Moreover, as the number of digits increases, the accuracy drops to approximately 41%.
Or in other words: they fail more than they succeed.
No need to mention that any of those calculations would have had a 100% success rate using a simple and plain calculator, right?
I leave you the link to
Keep reading with a 7-day free trial
Subscribe to Feasible to keep reading this post and get 7 days of free access to the full post archives.