AI

Google DeepMind makes AI history with gold medal win at world’s toughest math competition

July 22, 2025
0 6 minutes read

Do you want smarter insights into your inbox? Register for our weekly newsletters to get only what is important for Enterprise AI, data and security leaders. Subscribe now

Google DeepMind announced on Monday that an advanced version of his Gemini -artificial intelligence model has officially reached Gold medal level performance by the International mathematical OlympiadSolving five of the six exceptionally difficult problems and earning recognition as the first AI system that receives official assessment at the gold level of competition organizers.

The victory promotes the area of AI reasons and places Google ahead in the intensifying struggle between technical giants that build artificial intelligence of the next generation. What is even more important, it shows that AI can now tackle complex mathematical problems with the understanding of natural language instead of requiring specialized programming languages.

“Official results are in-gemini achieved gold medal level in the international mathematical Olympiad!” Demis HassabisCEO of Google DeepMind, wrote on the social media platform X on Monday morning. “An advanced version could solve 5 out of 6 problems. Incredible progress.”

The International mathematical OlympiadHeld annually since 1959, is generally regarded as the world’s most prestigious mathematics competition for pre-university students students. Each participating country sends six elite young mathematicians to compete in solving six exceptionally challenging problems of algebra, combinatorics, geometry and number theory. Only about 8% of human participants usually earn gold medals.

How Google DeepMind’s Gemini Deep thinks that Math’s most difficult problems have cracked

Google’s newest success is much more than the performance of 2024, when the company is combined Alphapro And Alphaageometry Systems earned the silver medal status by solving four of the six problems. That earlier system required human experts to first translate natural language problems into domain -specific programming languages and then interpret the mathematical output of the AI.

This year’s breakthrough came through Gemini Dieh DiepAn improved reasoning system that uses what researchers call “think in parallel. “In contrast to traditional AI models that follow a single reasoning chain, deep thinking is investigating several possible solutions at the same time before he arrives at a definitive answer.

“Our model worked end-to-end in natural language and produced rigorous mathematical evidence directly from the official problem descriptions,” Hassabis explained In a follow-up post on the social media site X, with which I emphasize that the system has completed its work within the standard 4.5-hour time limit of the competition.

The model reached 35 of a possible 42 points, which comfortably exceeds the gold medal threshold. According to IMO President Prof. dr. Dr. Gregor Dolinar were the solutions’Amazing in many ways“And turned out to be” clear, accurate and most of them easy to follow “due to competition prisoners.

See also  One week left to save up to $668 for Disrupt 2025 passes

OpenAi is confronted with recoil for circumventing official competitive rules

The announcement comes in the midst of a growing tension in the AI industry about competing practices and transparency. Google DeepMind’s measured approach to release the results of the AI community has praised, in particular in contrast to rival OpenAi’s handling of similar performance.

“We did not announce on Friday because we respected the original request from the IMO Board that all AI Laboratories only share their results after the official results were verified by independent experts and the students had rightly received the acclamation they deserved,” “” Hassabis wroteappear to the earlier announcement of OpenAi of his own Olympiad performance.

Social media users quickly noted the distinction. “Do you see? OpenAi ignored the IMO request. Shame. No class. Straight respect,” wrote a user. “Google DeepMind traded with integrity, tailored to humanity.”

The criticism stems from the decision of OpenAi to announce his own mathematical Olympiad results without participating in the official IMO evaluation process. Instead, OpenAi had a panel of former IMO participants the performance of its AI, an approach that some in the community image as a lack of credibility.

“OpenAi is perhaps the worst company on the planet at the moment,” wrote a critic, while others suggested that the company “should take things seriously” and “more credible.”

In the training methods that Gemini’s mathematical mastery have driven

The success of Google DeepMind seems to come from new training techniques that go beyond traditional approaches. The Team used advanced learning methods for reinforcement reinforcement designed to use multi-step reasoning, problem solution and racking data. The model also gained access to a composite collection of high-quality mathematical solutions and received specific guidelines for approaching IMO-style problems.

See also  Josh Allen's MVP necklace features white gold and 27 carats of diamonds

The technical performance impressed AI researchers who noticed his broader implications. “Not only solving mathematics … But understanding problems described by language and applying abstract logic to new cases,” wrote AI observer Elyss Wren. “This is not Rote’s memory – this is upcoming cognition in motion.”

Ethan MollickA professor at the Wharton School studying AI emphasized the importance of using a general model instead of specialized aids. “Increasing evidence of LLMS’s ability to generalize to new problem solving,” he wrote, and emphasized how this differs from previous approaches that required specialized mathematical software.

The model showed particularly impressive reasoning in one problem in which many human competitors applied mathematical concepts at graduate level. According to DeepMind researcher Junehyuk Jung, Gemini “made a brilliant observation and only used elementary numbers theory to create standing evidence”, finding a more elegant solution than many human participants.

What the victory of Google DeepMind means for the AI race of $ 200 billion

The breakthrough comes at a critical moment in the AI industry, where companies race to demonstrate superior reasoning options. The success has immediate practical implications: Google is planning to make a version of this DEPHINK MODEL Available for mathematicians for testing before rolling out to Google AI Ultra subscribers, who pay $ 250 every month for access to the company’s most advanced AI models.

The timing also emphasizes the intensifying competition between large AI laboratories. While Google celebrates its methodical, officially verified approach, the controversy about the announcement of OpenAI is reflected broader tensions about transparency and credibility in AI development.

See also  Visa just launched a protocol to secure the AI shopping boom — here’s what it means for merchants

This competitive dynamic extends beyond mathematical reasoning. In recent weeks, various AI companies have announced breakthrough options, although not all have been received positively. The Xai from Elon Musk has recently been launched Grock 4of which the company claimed that it was the ‘smartest AI in the world’, however Leaderboard -Scores showed it behind Behind models from Google and OpenAi. In addition, Grok has confronted with criticism of controversial characteristics, including Sexualized ai -companions and episodes of generating anti -Semitic.

The Dawn of AI that thinks as people with real consequences

The mathematical Olympiad victory goes beyond competitive. The performance of Gemini shows that AI systems can now match the reasoning at a human level in complex tasks that think creativity, abstractly and the ability to synthesize insights into several domains.

“This is an important progress compared to last year’s breakthrough result,” the DeepMind team noted In their technical announcement. The progression of requiring specialized formal languages to fully working in natural language suggests that AI systems become intuitive and accessible.

For companies, this development means that AI will soon be able to tackle complex analytical problems in various industries without specialized programming or domain expertise. The ability to reason through complicated challenges With the help of daily language, advanced analytical capacities between organizations can democratize.

Questions, however, continue to exist about whether these reasoning options will effectively translate into messier Real-World challenges. The mathematical Olympiad offers well-defined problems with clear success criteria removed from the ambiguous, versatile decisions that define the most business and scientific efforts.

Google DeepMind plans to return to next year’s competition “Looking for a perfect score. “The company believes that AI systems that combine fluency of natural language with rigorous reasoning” will become invaluable tools for mathematicians, scientists, engineers and researchers, which helps us to help human knowledge on the way to Agi. “

But perhaps the most meaningful details came from the competition itself: when the most difficult problem of the competition was confronted, Gemini started an incorrect hypothesis and he never recovered. Only five human students have correctly solved that problem. In the end it seems that even gold medal-winning AI still has something to learn from teenage wisions.


Source link
Tags
July 22, 2025
0 6 minutes read
Back to top button