OpenAI’s latest ChatGPT model, o3, has scored 136 on the Norwegian Mensa IQ test, surpassing 98% of human test-takers.
However, TechRadar reported that the Norwegian Mensa test is publicly available, raising concerns that o3 might have accessed questions or answers during its training. To address this, researchers at MaximumTruth.org developed a new, completely offline IQ test inaccessible to AI training data.
On this new test, designed to match the Mensa version’s difficulty, o3 scored 116, which is still impressive. This marks a significant leap from last year, when AI systems couldn’t break the 90-point barrier on similar tests.
As AI increasingly mimics human-like thinking, users are more prone to treating it as a conscious entity. A recent survey by EduBirdie revealed that 25% of Gen Z (born mid-1990s to early 2000s) already believe AI possesses consciousness, while over half expect AI to become conscious soon.