AI Does Math Using 1970s Typesetting Code — And That's a Warning

Here's something strange about the AI chatbots you may already use. When you ask one to solve a hard math problem, it doesn't quietly crunch numbers the way a calculator does. Instead, it writes out symbols like ∫, √ and fractions — and a blogger named Keenan Pepper spotted that this is TeX, a formatting code invented in the 1970s by Donald Knuth purely to make equations look pretty in printed books.

TeX was never meant to do math. It was meant to display it. Knuth would have found the idea funny. But AI models were trained on millions of documents full of TeX, so they learned the symbols — and, more importantly, learned what those symbols mean. Now the AI "thinks" by shuffling those symbols around, moving them like puzzle pieces until the answer looks right. That's a very different process from actually computing anything.

Why should you care? Because it explains a quirk you may have noticed: AI can explain advanced calculus beautifully and still get your grocery total wrong. It's pattern-matching text, not running calculations. The step-by-step reasoning usually isn't even shown to you. And when something truly hard is solved, like a famous unsolved physics equation, the trustworthy part is often a separate checking program — not the AI's own typing.

The practical takeaway is simple: treat AI math answers as a first draft, not a receipt. Great for learning, explaining, and getting unstuck. Not something to trust for your taxes, your loan payments, or a dosage calculation without checking. Also worth a moment of wonder — a tool built for printing books accidentally became the language AI uses to reason.