The quest to create strong systems that can match human intellect has seen a breakthrough in math capabilities as Alphabet’s Google and OpenAI announced that their artificial-intelligence models won gold medals at a worldwide mathematics tournament.
According to the findings, AI systems achieved the gold medal for the first time in the International Mathematical Olympiad for high-school students. Unlike the earlier methods employed by AI corporations, the models of both organizations were able to answer five of the six tasks. This was accomplished by utilizing general-purpose “reasoning” models that processed mathematical ideas using natural language.
According to Junehyuk Jung, a math professor at Brown University and visiting researcher in Google’s DeepMind AI unit, the accomplishment implies that mathematicians will start using AI in less than a year to address unresolved research challenges at the forefront of the discipline.
Jung told that he believes the possibility of cooperation between mathematicians and artificial intelligence will be made possible if we are able to answer challenging reasoning issues in natural language.
This achievement was made possible by a new experimental model developed by OpenAI that focused on dramatically scaling up “test-time compute.” This was accomplished by letting the model “think” for extended periods of time and using parallel processing power to execute several lines of reasoning at once, according to OpenAI researcher Noam Brown. OpenAI’s cost in terms of processing power was not disclosed by Brown, but he described it as “very expensive.”
For researchers at OpenAI, it is yet another clear indication that AI models are capable of sophisticated thinking that may extend beyond mathematics.
Google researchers share this confidence, believing that the skills of AI models may be applied to explore problems in other domains, like physics, according to Jung, who in 2003 earned an IMO gold medal as a student.
At the 66th IMO on the Sunshine Coast in Queensland, Australia, 67 competitors, or almost 11% of the 630 pupils, won gold medals.
Using math-specific AI algorithms, Google’s DeepMind AI unit won a silver medal last year. Google employed a general-purpose model this year called Gemini Deep Think, which was first introduced at its annual developer conference in May.
This year’s Google method worked exclusively in natural language and resolved the issues within the allotted 4.5 hours, the firm stated in a blog post, in contrast to earlier AI initiatives that depended on formal languages and time-consuming processing.
According to a post by researcher Alexander Wei on social networking site X, OpenAI, which has its own collection of reasoning models, also created an experimental version for the competition. “It will be several months before the company releases anything with this level of math capability,” he said.
This year represented the competition’s first official collaboration with certain AI developers, who have long utilized well-known math challenges like IMO to evaluate model skills. IMO judges confirmed the results of those businesses, including Google, and directed them to publish them on July 28.
On Monday, Google DeepMind CEO Demis Hassabis stated on X, “We honored the IMO Board’s initial request that all AI labs share their results only after the official results had been confirmed by independent experts and the students had rightfully received the acclaim they deserved.”
OpenAI, which revealed its findings on Saturday and claimed gold medal status for the first time, said in an interview that it obtained authorization from an IMO board member to do so after the Saturday closing ceremony.
On Monday, IMO board president Gregor Dolinar told that the competition permitted cooperating firms to release their findings.
