Why Most AIs Are So Dumb

It’s because they are, according to IQ tests. Here are the charts.

Hey, AI! You’re repeating yourself as you gaslight me. Cool cool cool. (Photo by Kin Shing Lai on Unsplash)

Have you ever felt like talking to an AI chatbot is like talking to the slowest kid in class, on a bad day?

There is a good reason for that.

It’s because most Large Language Models (LLMs) are, in fact, dumb.

Yes, even the “thinking” models have IQs lower than you’d expect. And thanks to the fine folks over at TrackingAI.org, I have the proof.

Screenshot of TrackingAI.org’s homepage by the author, showing IQ Test Results for “Offline Test” for AIs.

Claude Sonnet 3.7 (from Anthropic) is the red-beige “thinking person” icon covering up ChatGPT o1 (from OpenAI).

Have you, like me, been surprised by Gemini Pro 2.5 (from Google)’s abilities? Particularly for coding it seems the best to me at the moment.

Remember how awful Bard was? It was literally so bad Google changed the name of the product to its current name, Gemini.

Ugh, I still cringe remembering it.

Bard was truly not capable of anything useful, and Google no longer offers it in any form; now it’s all Gemini.