IQ of artificial intelligence. Collages: Ion Mateș / Hotnews. Photo: Dreamstime
The new Chatgpt model, version O3, has just obtained an IQ score of 136 in the Mensa test administered in Norway-higher than 98% of humanity, given that artificial intelligence models (AI) have become much more complex, more flexible and, in a way, more intelligent, within one year, writes Techradar.
The progress made by AI have been so spectacular that some young people believe that it becomes self -aware.
According to a new poll conducted by the online Edubirdie platform, 25% of the Z -generation representatives now believe that artificial intelligence is already self -aware, and more than half I think it is only a matter of time until their chatbot becomes aware and possibly request the right to vote.
However, there is an important context to take into account the IQ test. The Mensa test in Norway is public, which means that, from a technical point of view, the model may have been trained beforehand with its questions or answers.
So the researchers at the MaximTrutruth.org site have created a new IQ test that is completely “offline” and inaccessible for data training.
In this test, designed to have a degree of difficulty equivalent to that of the Mensa version, the O3 model obtained a score of 116, also very high.
This result would place the O3 of Openai model in the top 15% of human intelligence, if it were a person, and not an artificial intelligence. For example, the data published by WorldPopopulationReview shows that the average IQ of the population would be 99.7 in the United States, 102 in Canada, 106 in Japan, Taiwan and Singapore, and 96.3 in Romania.
Last year, no model Ai got a score of over 90 at the same test. In May last year, the best performing you have difficulties with the rotated triangles of tests. Now, the O3 model is comfortable on the right side of Gauss's curve in terms of intelligence, along with the smartest people.
And the area in the curve becomes more and more crowded. Claude, the chatbot developed by Anthropic, has climbed a little bit. Gemini, the chatbot created by Google, has obtained scores in area 90. Even GPT-4o, the default model of Chatgpt, is only a few IQ points under O3.
As the journalists from Techradar emphasizes, it is not just that these models become smarter. It is about learning quickly and that they improve as software does, not as people do.
Photo article: Ai / Rokas Tenys / Dreamstime.com.
