AI cracks math Olympiad: Google, OpenAI models win gold at global competition
The results marked the first time that AI systems crossed the gold-medal scoring threshold at the International Mathematical Olympiad for high-school students. Both companies' models solved five out of six problems, achieving the result using general-purpose 'reasoning' models that processed mathematical concepts using natural language, in contrast to the previous approaches used by AI firms.
The achievement suggests AI is less than a year away from being used by mathematicians to crack unsolved research problems at the frontier of the field, according to Junehyuk Jung, a math professor at Brown University and visiting researcher in Google's DeepMind AI unit.
'I think the moment we can solve hard reasoning problems in natural language will enable the potential for collaboration between AI and mathematicians,' Jung told Reuters.
OpenAI's breakthrough was achieved with a new experimental model centred on massively scaling up 'test-time compute.'
This was done by both allowing the model to 'think' for longer periods and deploying parallel computing power to run numerous lines of reasoning simultaneously, according to Noam Brown, researcher at OpenAI.
Brown declined to say how much in computing power it cost OpenAI, but called it 'very expensive.'
To OpenAI researchers, it is another clear sign that AI models can command extensive reasoning capabilities that could expand into other areas beyond math.
The optimism is shared by Google researchers, who believe AI models' capabilities can apply to research quandaries in other fields such as physics, said Jung, who won an IMO gold medal as a student in 2003.
Of the 630 students participating in the 66th IMO on the Sunshine Coast in Queensland, Australia, 67 contestants, or about 11 per cent, achieved gold-medal scores.
Google's DeepMind AI unit last year achieved a silver medal score using AI systems specialised for math. This year, Google used a general-purpose model called Gemini Deep Think, a version of which was previously unveiled at its annual developer conference in May.
Unlike previous AI attempts that relied on formal languages and lengthy computation, Google's approach this year operated entirely in natural language and solved the problems within the official 4.5-hour time limit, the company said in a blog post.
OpenAI, which has its own set of reasoning models, similarly built an experimental version for the competition, according to a post by researcher Alexander Wei on social media platform X. He noted that the company does not plan to release anything with this level of math capability for several months.
This year marked the first time the competition coordinated officially with some AI developers, who have for years used prominent math competitions like IMO to test model capabilities. IMO judges certified the results of those companies, including Google, and asked them to publish results on July 28.
'We respected the IMO Board's original request that all AI labs share their results only after the official results had been verified by independent experts and the students had rightly received the acclamation they deserved,' Google DeepMind CEO Demis Hassabis said on X on Monday.
OpenAI, which published its results on Saturday and first claimed gold-medal status, said in an interview that it had permission from an IMO board member to do so after the closing ceremony on Saturday.
The competition on Monday allowed cooperating companies to publish results, Gregor Dolinar, president of IMO's board, told Reuters. — Reuters
Hashtags
Try Our AI Features
Explore what Daily8 AI can do for you:
Comments
No comments yet...
Related Articles
The Star
3 hours ago
- The Star
Trump's order to block 'woke' AI in government encourages tech giants to censor their chatbots
Tech companies looking to sell their artificial intelligence technology to the US federal government must now contend with a new regulatory hurdle: prove their chatbots aren't "woke.' President Donald Trump's sweeping new plan to counter China in achieving "global dominance' in AI promises to cut regulations and cement American values into the AI tools increasingly used at work and home. But one of Trump's three AI executive orders signed July 23 – the one "preventing woke AI in the federal government' – marks the first time the US government has explicitly tried to shape the ideological behavior of AI. Several leading providers of the AI language models targeted by the order – products like Google's Gemini and Microsoft's Copilot – have so far been silent on Trump's anti-woke directive, which still faces a study period before it gets into official procurement rules. While the tech industry has largely welcomed Trump's broader AI plans, the anti-woke order forces the industry to leap into a culture war battle – or try their best to quietly avoid it. "It will have massive influence in the industry right now,' especially as tech companies are already capitulating to other Trump administration directives, said civil rights advocate Alejandra Montoya-Boyer, senior director of The Leadership Conference's Center for Civil Rights and Technology. The move also pushes the tech industry to abandon years of work to combat the pervasive forms of racial and gender bias that studies and real-world examples have shown to be baked into AI systems. "First off, there's no such thing as woke AI,' Montoya-Boyer said. "There's AI technology that discriminates and then there's AI technology that actually works for all people.' Moulding the behaviours of AI large language models is challenging because of the way they're built and the inherent randomness of what they produce. They've been trained on most of what's on the internet, reflecting the biases of all the people who've posted commentary, edited a Wikipedia entry or shared images online. "This will be extremely difficult for tech companies to comply with,' said former Biden official Jim Secreto, who was deputy chief of staff to U.S. Secretary of Commerce Gina Raimondo, an architect of many of Biden's AI industry initiatives. "Large language models reflect the data they're trained on, including all the contradictions and biases in human language.' Tech workers also have a say in how they're designed, from the global workforce of annotators who check their responses to the Silicon Valley engineers who craft the instructions for how they interact with people. Trump's order targets those "top-down' efforts at tech companies to incorporate what it calls the "destructive' ideology of diversity, equity and inclusion into AI models, including "concepts like critical race theory, transgenderism, unconscious bias, intersectionality, and systemic racism.' For Secreto, the order resembles China's playbook in "using the power of the state to stamp out what it sees as disfavored viewpoints." The method is different, with China relying on direct regulation through its Cyberspace Administration, which audits AI models, approves them before they are deployed and requires them to filter out banned content such as the bloody Tiananmen Square crackdown on pro-democracy protests in 1989. Trump's order doesn't call for any such filters, relying on tech companies to instead show that their technology is ideologically neutral by disclosing some of the internal policies that guide the chatbots. "The Trump administration is taking a softer but still coercive route by using federal contracts as leverage,' Secreto said. "That creates strong pressure for companies to self-censor in order to stay in the government's good graces and keep the money flowing.' The order's call for "truth-seeking' AI echoes the language of the president's one-time ally and adviser Elon Musk, who frequently uses that phrase as the mission for the Grok chatbot made by his company xAI. But whether Grok or its rivals will be favored under the new policy remains to be seen. Despite a "rhetorically pointed' introduction laying out the Trump administration's problems with DEI, the actual language of the order's directives shouldn't be hard for tech companies to comply with, said Neil Chilson, a Republican former chief technologist for the Federal Trade Commission. "It doesn't even prohibit an ideological agenda,' just that any intentional methods to guide the model be disclosed, said Chilson, who is now head of AI policy at the nonprofit Abundance Institute. "Which is pretty light touch, frankly.' Chilson disputes comparisons to China's cruder modes of AI censorship. "There is nothing in this order that says that companies have to produce or cannot produce certain types of output,' he said. "It says developers shall not intentionally encode partisan or ideological judgments. That's the exact opposite of the Chinese requirement.' So far, tech companies that have praised Trump's broader AI plans haven't said much about the order. OpenAI on Thursday said it is awaiting more detailed guidance but believes its work to make ChatGPT objective already makes the technology consistent with what the order requires. Microsoft, a major supplier of email, cloud computing and other online services to the federal government, declined to comment Thursday. Musk's xAI, through spokesperson Katie Miller, a former Trump official, pointed to a company comment praising Trump's AI announcements as a "positive step' but didn't respond to a follow-up question about how Grok would be affected. xAI recently announced it was awarded a U.S. defense contract for up to US$200mil, just days after Grok publicly posted a barrage of antisemitic commentary that praised Adolf Hitler. Anthropic, Google, Meta, and Palantir didn't immediately respond to emailed requests for comment Thursday. AI tools are already widely used in the federal government, including AI platforms such as ChatGPT and Google Gemini for internal agency support to summarize the key points of a lengthy report. The ideas behind the order have bubbled up for more than a year on the podcasts and social media feeds of Trump's top AI adviser David Sacks and other influential Silicon Valley venture capitalists, many of whom endorsed Trump's presidential campaign last year. Much of their ire centered on Google's February 2024 release of an AI image-generating tool that produced historically inaccurate images before the tech giant took down and fixed the product. Google later explained that the errors – including one user's request for American Founding Fathers that generated portraits of Black, Asian and Native American men – were the result of an overcompensation for technology that, left to its own devices, was prone to favoring lighter-skinned people because of pervasive bias in the systems. Trump allies alleged that Google engineers were hard-coding their own social agenda into the product, and made it a priority to do something about it. "It's 100% intentional,' said prominent venture capitalist and Trump adviser Marc Andreessen on a podcast in December. "That's how you get Black George Washington at Google. There's override in the system that basically says, literally, 'Everybody has to be Black.' Boom. There's squads, large sets of people, at these companies who determine these policies and write them down and encode them into these systems.' Sacks credited a conservative strategist who has fought DEI initiatives at colleges and workplaces for helping to draft the order. "When they asked me how to define 'woke,' I said there's only one person to call: Chris Rufo. And now it's law: the federal government will not be buying WokeAI,' Sacks wrote on X. Rufo responded that, in addition to helping define the phrase, he also helped "identify DEI ideologies within the operating constitutions of these systems.' – AP
The Star
4 hours ago
- The Star
Intel's foundry future depends on securing a customer for next-gen chipmaking tech
FILE PHOTO: Intel logo is seen near computer motherboard in this illustration taken January 8, 2024. REUTERS/Dado Ruvic/Illustration/ File Photo SAN FRANCISCO (Reuters) -Intel warned investors on Thursday that it may have to get out of the chip manufacturing business if it does not land external customers to make chips in its factories. New CEO Lip-Bu Tan said on Thursday the company's engineers were busy working with customers to jump-start its next-generation contract manufacturing process, or foundry, as the company announced big layoffs alongside a wider-than-expected third-quarter loss outlook. Those customers for the company's so-called 14A manufacturing process are crucial to the success of the technology - so much so that if it fails to secure a big one, it could shut down its cutting-edge manufacturing business altogether, according to Intel's quarterly filing onThursday. The possibility that Intel could drop out of the cutting-edge manufacturing business would be a historic shift for a company that has described itself as a steward of Moore's Law - an observation by Intel co-founder Gordon Moore about the fast rate of development of the chip industry that held true for decades. Intel is the only U.S. chipmaker capable of making advanced computing chips. Intel has struggled for years due to management missteps, missing out on the AI race and losing market share to its longtime rival AMD. Former CEO Patrick Gelsinger poured money into Intel's foundry business, aiming to compete with chip manufacturing giant TSMC. Tan, who has already taken steps to right the ship, said on a post-earnings call on Thursday that he was personally reviewing all chip designs and investments. "We're developing Intel 14A ... from the ground up in close partnership with large external customers," Tan said in a memo released with the results. "Going forward, our investment in Intel 14A will be based on confirmed customer commitments. "We will build what our customers need, when they need it, and earn their trust through consistent execution." Intel said that without a significant customer, it would consider cancelling or pausing development of 14A and subsequent technologies. Should the company take the step, it planned to continue to manufacture chips with its 18A technology and a variant through 2030, according to the filing. In a post-earnings conference call, Tan said on Thursday that he is focused on working with customers to ensure 14A is a success and that tight collaboration with external customers is something that was absent from the company's 18A, which is set to go into high-volume production later this year. Tan said bringing those prospective customers in and gaining their feedback during 14A's development has already made it more promising than 18A. "That gave me a lot more confidence that this time, we have customers (that) are engaging early enough in the inception" of 14A, Tan said. "We learn from our mistakes, and we can learn quicker and then get a better result." The consequences of a decision to halt internal manufacturingwould be significant for Intel, the filing said. It would mean that over time, Intel would become dependent on Taiwan's TSMC for contract manufacturing, or foundry, services. Doing so would also put it at a competitive disadvantage to competitors such as AMD, which has longer relationships and experience working with TSMC. Intel had roughly $100 billion of chipmaking equipment as of June 28. If the company halted its 14A manufacturing line, the company expects "significant material impairments" related to the company's foundry assets, the company's filing said. (Reporting by Max A. Cherney and Stephen Nellis in San Francisco; Additional reporting by Arsheeya Bajwa in Bengaluru; Editing by Sayantani Ghosh and Stephen Coates)
Free Malaysia Today
5 hours ago
- Free Malaysia Today
Intel revenue climbs as 15% workforce cut takes effect
Intel, once a Silicon Valley icon, has been overshadowed by TSMC and Samsung, which dominate the custom semiconductor manufacturing sector. (AFP pic) SAN FRANCISCO : Intel on Thursday posted quarterly revenue that topped market expectations, saying it has cut about 15% of its workforce to be 'more agile.' The US chip maker also said it 'will no longer move forward' with projects in Germany and Poland as part of a push to save billions of dollars. The struggling chip maker's earnings report came as rivals specialising in graphics processing units (GPUs) for artificial intelligence thrive due to rapid adoption of the technology. Intel is one of Silicon Valley's most iconic companies, but its fortunes have been dwarfed by Asian powerhouses TSMC and Samsung, which dominate the made-to-order semiconductor business. The company was also caught by surprise with the emergence of Nvidia as the world's preeminent AI chip provider. Intel's niche has been in chips used in traditional computing processes, steadily being eclipsed by the AI revolution. Intel reported US$12.9 billion in sales in the recently ended quarter, topping forecasts, but logged a US$2.9 billion loss that included US$1.9 billion in restructuring charges. 'Intel has completed the majority of the planned headcount actions it announced last quarter to reduce its core workforce by approximately 15%,' the company said in an earnings release. 'These changes are designed to create a faster-moving, flatter and more agile organisation.' Intel shares were down slightly in after-hours trades that followed the release of the earnings figures. Intel chief executive Lip-Bu Tan took the helm in March, announcing layoffs as White House tariffs and export restrictions muddied the market. Malaysia-born tech industry veteran Tan has said it 'won't be easy' to overcome challenges faced by the company. Demand and turmoil Meanwhile, South Korean chip giant SK hynix reported record quarterly profits Thursday thanks to soaring demand for artificial intelligence technology. The world's second-largest memory chip maker dominates the market for high-bandwidth memory semiconductors and is a key supplier for US titan Nvidia. Riding the AI wave, last week Taiwan chip giant TSMC announced a surge in net profit for the second quarter. 'Nvidia suppliers like SK hynix will continue to enjoy strong demand in the coming months and years for memory chips due to the high memory content needed to make AI chips functional,' G. Dan Hutcheson of TechInsights told AFP. Dutch tech giant ASML last week said it booked higher net profits in the second quarter of 2025 compared with the same period last year. The firm, which makes cutting-edge machines for the manufacture of semiconductors, warned that the growth outlook for next year was somewhat less rosy than before. 'Looking at 2026, we see that our AI customers' fundamentals remain strong,' said CEO Christophe Fouquet in a statement. 'At the same time, we continue to see increasing uncertainty driven by macro-economic and geopolitical developments,' he cautioned. Washington has sought to curb exports of state-of-the-art chips to China, concerned that they could be used to advance Beijing's military systems and otherwise undermine American dominance in AI.
