AI Does Multiplication Underneath. So Why Did Older Models Break at School Maths?
Like anyone who got pulled into the AI wave, I went through the whole thing. The amazement phase, the over reliance phase, the okay this thing just hallucinated phase.., and then the slow ohh okay I think I am starting to get what this actu
AI Does Multiplication Underneath. So Why Did Older Models Break at School Maths?

Image generated by Chatgpt
Like anyone who got pulled into the AI wave, I went through the whole thing. The amazement phase, the over reliance phase, the okay this thing just hallucinated phase.., and then the slow ohh okay I think I am starting to get what this actually is phase.
Ikr… everyone is talking about MCP, agents, multi agent workflows, RAG pipelines, and whatever dropped on HuggingFace this morning. And I will get there. But today I just want to sit with a small problem that nobody really talks about anymore because it feels too old to bring it up. A schoolboy problem. The kind where your maths teacher would give you that look.
Somewhere in the middle of all that AI chaos, I came across the 9.9 versus 9.11 meme.

Someone asked a model which number is bigger, and it confidently picked 9.11 because 11 is bigger than 9. Bro. This is not a version number. This is not some edge case in floating point arithmetic. This is the decimal system. The one we all learned before we learned anything else.
To be fair, most good models get this right today. I knew this one is old. But that is kind of the point, the mistake itself was never what was interesting. What was interesting was the pattern underneath it.
The model does not look confused when it is wrong. It does not slow down, add a disclaimer, or pause like someone who just remembered the exam has negative marking. It gives the answer with the full energy of a person who has already submitted the paper, walked out of the hall, and is thinking about lunch. Then you correct it. It apologises beautifully. Very polite, very composed, basically a customer support email. And then sometimes, with slightly different numbers, same mistake from a different angle.
That is what stayed with me. This is the same model that can explain transformers, summarise long documents, write decent code, and give surprisingly calm life advice at 1AM when your own brain has completely checked out. How does something that capable not know which decimal is larger? And more than that, why does it not know that it does not know?
A small signal for the next reader
Did this help you understand the idea?
of readers found this useful
Thanks, your signal was saved.