LLMs cannot count... not without using code at least

A pure LLM will only take its best guess at the answer to your question. That means that if it's supposed to calculate, it will probably get it wrong. UNLESS it uses code.

So, when asking your LLM to analyze data or calculate something, make sure you put the phrase "use code" into the prompt. That's it!

An AI developer sold me that trick, and today I saw it in action for the first ime, when working with Copilot + activate ChatGPT 5. Without this phrase it was not able to calculate, adding "Use code" made all the difference.

Try it and let me know :-)

6 comments