AI may pass the world's most difficult scientific test by the end of the year. Thus, artificial intelligence could become smarter than most scientists.
Experts believe that AI will be able to pass the world's most difficult scientific test called "The Final Exam of Humanity" with a 100% score. This could happen in just a few months or within a year, writes Focus.
In 2024, researchers from Scale AI (which specializes in data labeling necessary for AI training) and the non-profit organization Center for AI Safety collected 70,000 questions from scientists in 50 countries to create the world's most difficult scientific test.
As a result, a test called "The Final Exam of Humanity" was created, consisting of 2,500 questions on mathematics, humanities, and natural sciences. This test is used to assess the knowledge level of AI.
Each question in the test requires at least a doctoral degree in a particular field of science. Answering all questions correctly can earn the title of "universal expert."
The test was designed to study both the breadth of AI's knowledge and the depth of its reasoning. The questions require short and unambiguous answers, and finding answers online is challenging. Thus, AI must possess a very broad range of knowledge to pass this test.
In 2024, OpenAI's ChatGPT scored only 3% on the test, while Google's Gemini and Anthropic's Claude showed slightly better results.
At that time, this test helped dispel concerns that AI could become too intelligent and demonstrated a significant knowledge gap between large language models and the world's best scientists.
However, recently, the large language model Gemini achieved a score of 45.9% on "The Final Exam of Humanity," while Claude reached a score of 34.2%. ChatGPT is not far behind its competitors.
Calvin Zhang from Scale AI stated that AI learns very quickly and could achieve a perfect score on "The Final Exam of Humanity" in just a few months or within a year. This means that if AI passes the world's most difficult scientific test, future artificial intelligence will need to be tested on questions that no human knows the answer to.
Still, Zhang believes that AI will not be able to fully master certain narrow fields of science and will find it very difficult to compete with humans in decision-making, including judgment and creativity.
