None of the neural networks failed the test developed by the organization “Center for Artificial Intelligence Security”.
What else is known
This exam, called “the last test of humanity”, is characterized by high complexity and requires AI to answer several thousand questions from disciplines such as physics, mathematics, chemistry and other scientific fields. Considerable complexity for neural networks is created by numerous auxiliary materials that need to be analyzed to find the correct answers.
The creators of the test emphasize that not every person can cope with it, because the questions are based on materials from advanced university courses. The best result among AIs was demonstrated by the updated version of ChatGPT, which answered 8.3% of the questions, while the total number of correct answers among all systems did not exceed 10%.