OpenAI’s Latest Chatbot Aces IQ Test, Now Smarter than 9 out of 10 People
OpenAI’s latest model o1 just passed the Norwegian Mensa IQ test, achieving a score that doesn’t just surpass other AI models but also most human results.
Tested by TrackingAI.org’s Maxim Lott (read his analysis here), the model correctly answered 25 out of 35 IQ questions. This marks a substantial improvement over previous AI models, which had shown little progress in IQ scores since earlier this year.
To validate these results and address concerns about potential training data bias, Lott created a new set of IQ questions that were kept offline and inaccessible to AI training algorithms. Even on this novel test, o1 maintained its significant lead over other AI models, suggesting a genuine advancement in reasoning capabilities rather than mere memorization of existing test data.
The model demonstrated its prowess by solving complex pattern recognition problems and providing detailed explanations for its answers. In one instance, o1 correctly identified a subtle pattern in a challenging visual puzzle, offering insights that rivaled human expert analysis.
While o1 still makes occasional errors, its performance on both established and newly created IQ tests indicates that AI reasoning abilities are rapidly approaching human-level intelligence in certain domains. The results put estimates of the current AI IQ to be between 100 and 120.
From this, Lott makes a huge projection: “We should start seeing AIs breaking 140 IQ in 2026.”
That’s at least 20 whole points in just two years. After what we’ve seen so far, with barely two years since ChatGPT was first launched, is it time to ask again, is the world ready for actually intelligent AI?
Information for this story was found via the sources and companies mentioned. The author has no securities or affiliations related to the organizations discussed. Not a recommendation to buy or sell. Always do additional research and consult a professional before purchasing a security. The author holds no licenses.
