Terence Tao supports it! AI Mathematical Olympiad Progress Award Announced, with Prizes of More than US$1 Million

How can a large model do math problems? You might as well let it participate in this level of competition. "Preliminary results for the AI ​​Mathematical Olympiad (AIMO Progress Award) have been announced. According to the leaderboard data, it currently appears that the winning program correctly answered 29/50 questions in the private test, which is a higher than expected result. "Just now, Tao Zhexuan announced the news on his personal blog.

AIMO originated from XTX Markets, a non-bank financial institution that specializes in machine learning quantitative trading. It mainly allows participants to use AI models to solve international mathematical problems. The goal of this progress award is to create a system that can solve complex mathematical problems written in LaTeX format. algorithms and models. This helps to promote the mathematical reasoning capabilities of artificial intelligence models and promote the development of cutting-edge knowledge.
It is worth mentioning that this is the first time that the Progress Award will be awarded, and the winning team will compete for a prize of US$1.048 million. The participating questions include questions from intermediate high school mathematics competitions, but the difficulty level does not reach the level of the International Mathematical Olympiad (IMO). Eligibility for the award requires teams to publicly release their code, methods, data, and model parameters.
According to the rankings, we can see that Numina’s team won first place, CMU_MATH ranked second, after exams temporarily ranked third, and codeinter and Conor #2 teams obtained fourth and fifth results respectively. According to the rules, the maximum number of participating teams is 5 people, and two of the top five are single players.

The models used by everyone are also different, including Mixtral・8x7b, Gemma, Llama 3, etc.

1. The IMO International Mathematical Olympiad has always been directly participated by enthusiasts who specialize in mathematics. However, as the capabilities of large models continue to improve, their performance in mathematics has received more and more attention.

2. The focus of everyone’s testing has gradually shifted from similar college entrance examination questions to the Olympic arena. Previously, AlphaGeometry from Google DeepMind was published in Nature. This research solved 25 of the 30 Olympic-level problems, approaching the International Mathematical Olympiad. Average performance of competition gold medalists.
3. With the explosion of large models, more than one company has begun to challenge the Olympiad problem. What will be the result if they compete with each other?
4. Last November, a mathematics competition specifically set up for AI came with a prize of US$10 million, aiming to stimulate the development of open artificial intelligence models so that they can perform with top humans in the International Mathematical Olympiad (IMO) are equally outstanding. The Progress Award is awarded as part of the AIMO Awards.

   

   AIMO’s advisory committee members are all big names in mathematics, including Fields Medal winners Timothy Gowers and Terence Tao, and they serve together with Po-Shen Loh, Dan Roberts and Geoff Smith.

   

   Reference link: https://aimoprize.com/participate

The above is the detailed content of Terence Tao supports it! AI Mathematical Olympiad Progress Award Announced, with Prizes of More than US$1 Million. For more information, please follow other related articles on the PHP Chinese website!