Smarter than most people. The new OpenAI AI model showed a stunning result in an authoritative IQ test

“Smarter than most people. OpenAI’s new AI model showed a stunning result in the authoritative IQ test September 17, 09:35 Share: o1 correctly answered 25 out of 35 questions of the IQ test (Photo: OpenAI) Author: Anastasiya Pechenyuk OpenAI’s artificial intelligence model o1 surpassed all competitors in the authoritative Norwegian Mensa IQ test. Model o1 correctly answered 25 out of 35 IQ test questions. It used questions that had never been published”, — write on: ua.news

Model o1 correctly answered 25 out of 35 IQ test questions. It used questions that had never been published on the Internet, which search engines have access to. This result allowed the model to score 120 IQ points. For comparison, Anthropic’s Claude AI model could achieve such a result in only 1.5-2.5 years.

“Claude-1 … gave 6 correct answers, giving an IQ of about 64. Claude-2 scored 6 additional points for the test (about 18 IQ points). Claude-3 scored another 6.5 points, which is worth about 19 more IQ points, and which is above the human average. … A very simple extrapolation suggests that we should expect Claude-4 in 12-16 months, and that it will have to answer about 25 questions per test,” writes artificial intelligence researcher Max Lott, who tested the new model.

The most difficult test solved by o1 involves pattern analysis and the ability to identify missing elements in the grid. The model demonstrated improved understanding of geometric patterns. Despite some missteps on similar FAQs, o1 showed that OpenAI has made significant progress in the development of high-performance AI. The result of the test also suggests that in the coming years AI will exceed the mark of 140 IQ points.