Humans beat AI gold-level score at top maths contest
Neither model scored full marks — unlike five young people at the International Mathematical Olympiad (IMO), a prestigious annual competition where participants must be under 20 years old.
Google said Monday that an advanced version of its Gemini chatbot had solved five out of the six maths problems set at the IMO, held in Australia's Queensland this month.
'We can confirm that Google DeepMind has reached the much-desired milestone, earning 35 out of a possible 42 points — a gold medal score,' the US tech giant cited IMO president Gregor Dolinar as saying.
'Their solutions were astonishing in many respects. IMO graders found them to be clear, precise and most of them easy to follow.'
Around 10 percent of human contestants won gold-level medals, and five received perfect scores of 42 points.
US ChatGPT maker OpenAI said that its experimental reasoning model had scored a gold-level 35 points on the test.
The result 'achieved a longstanding grand challenge in AI' at 'the world's most prestigious math competition', OpenAI researcher Alexander Wei wrote on social media.
'We evaluated our models on the 2025 IMO problems under the same rules as human contestants,' he said.
'For each problem, three former IMO medalists independently graded the model's submitted proof.'
Google achieved a silver-medal score at last year's IMO in the British city of Bath, solving four of the six problems.
That took two to three days of computation — far longer than this year, when its Gemini model solved the problems within the 4.5-hour time limit, it said.
The IMO said tech companies had 'privately tested closed-source AI models on this year's problems', the same ones faced by 641 competing students from 112 countries.
Hashtags
Try Our AI Features
Explore what Daily8 AI can do for you:
Comments
No comments yet...
Related Articles
Express Tribune
13 hours ago
- Express Tribune
Genie 3 by Google brings AI closer to AGI with realistic world-building
Google DeepMind has launched Genie 3, an advanced AI world model, claiming it represents a crucial step towards artificial general intelligence (AGI). The foundation model, still in a research preview, is designed to generate interactive 3D environments in real time, a significant advancement over previous models. Genie 3, which was announced through a blogpost on Google's website, can produce several minutes of photo-realistic simulations and maintain physical consistency across scenarios, learning from past generated outputs to enhance its world-building. What if you could not only watch a generated video, but explore it too? 🌐 Genie 3 is our groundbreaking world model that creates interactive, playable environments from a single text prompt. From photorealistic landscapes to fantasy realms, the possibilities are endless. 🧵 — Google DeepMind (@GoogleDeepMind) August 5, 2025 Unlike its predecessor, Genie 2, which could only generate short simulations, Genie 3's capabilities extend to creating complex, interactive virtual worlds with dynamic, long-term consistency. This allows for more accurate simulations of real-world physics, moving towards training AI agents for general-purpose tasks, essential for AGI development. DeepMind sees the model as a game-changer in training embodied agents, whose real-world interactions are particularly challenging to simulate. With Genie 3, AI agents are expected to learn by interacting with and adapting to their environments, much like humans do in real life. One nice thing you can do with an interactive world model, look down and see your footwear ... and if the model understands what puddles are. Genie 3 creation. — Matt McGill (@MattMcGill_) August 5, 2025 This self-learning approach is seen as vital for advancing AGI, pushing AI towards human-like intelligence. Despite its potential, the model has limitations, including difficulty modelling complex interactions between agents and the limited duration of continuous interactions. However, it marks an important development in the journey to AGI, offering a glimpse into a future where AI can plan, explore, and improve autonomously through trial and error. DeepMind's Genie 3, alongside its previous models, represents a leap forward in AI's ability to interact with the world and learn from experience. Genie 3 feels like a watershed moment for world models 🌐: we can now generate multi-minute, real-time interactive simulations of any imaginable world. This could be the key missing piece for embodied AGI… and it can also create beautiful beaches with my dog, playable real time — Jack Parker-Holder (@jparkerholder) August 5, 2025
Express Tribune
14 hours ago
- Express Tribune
WhatsApp introduces features to spot scams
WhatsApp has announced the launch of new features aimed at protecting users from scams, as the company steps up its efforts to combat online fraud. These new tools, designed to identify potential scams in both group and individual chats, come alongside the removal of over 6.8 million accounts linked to criminal scam operations globally. The updated features, announced on Meta's website, aim to empower users with better tools to detect and prevent scams. One of the key updates is the introduction of a safety overview for group chats. This feature will notify users when they are added to a group by someone who isn't in their contact list. The safety overview will display essential information about the group, including whether any members are contacts of the user. added to a group you don't recognize? 🧐 if that happens, we give you info about the group and suggest safety tools you can use to decide if it's a group you want to stay in or leave — WhatsApp (@WhatsApp) August 5, 2025 This added context aims to help users make informed decisions before engaging in the group chat. If a user chooses to continue exploring the group, they can view more context on the chat, with notifications muted until they decide whether to stay or leave. In addition to improving group chat security, WhatsApp is addressing scams in private messaging. Scammers often try to initiate conversations through other platforms, only to later divert the user to WhatsApp. In response, WhatsApp is testing features that alert users when they begin a chat with someone outside their contact list. These notifications will provide additional context on the person they are messaging, helping users make more informed decisions before engaging in the conversation. WhatsApp has also partnered with OpenAI to disrupt scam operations, specifically targeting a scam centre in Cambodia. The scammers had used ChatGPT to generate deceptive messages that lured people into schemes involving fake likes, pyramid schemes, and cryptocurrency investments. The scams often involved directing victims to Telegram, where they were tasked with liking TikTok videos and then asked to invest in cryptocurrency. WhatsApp's safety guidelines encourage users to take a moment before responding to unfamiliar messages. Meta on Tuesday said it shut nearly seven million WhatsApp accounts linked to scammers in the first half of this year and is ramping up safeguards against such schemes. — AFP News Agency (@AFP) August 5, 2025 They advise users to assess whether the message appears legitimate, question the urgency of the request, and verify the identity of anyone claiming to be a friend or family member through other communication methods. By introducing these features, WhatsApp is taking significant steps to protect its users from scams and reinforce the platform's commitment to online safety.
Express Tribune
14 hours ago
- Express Tribune
OpenAI launches advanced open-source AI reasoning models for developers
OpenAI has unveiled two open-source AI reasoning models, gpt-oss-120b and gpt-oss-20b, marking a significant shift in the company's approach to AI. The models, announced on OpenAI's website, are now available for free download from the Hugging Face platform. The models come in two sizes: a larger version designed for high-powered systems and a lighter version suited for consumer-grade laptops. These models, which build upon the capabilities of OpenAI's proprietary o-series, are the first open-weight models released since the launch of GPT-2. We released two open-weight reasoning models—gpt-oss-120b and gpt-oss-20b—under an Apache 2.0 license. Developed with open-source community feedback, these models deliver meaningful advancements in both reasoning capabilities & — OpenAI (@OpenAI) August 5, 2025 The new models employ cutting-edge reasoning techniques that allow them to tackle complex queries and process large amounts of information in parallel, leveraging multiple AI agents for better results. OpenAI has trained the models using high-compute reinforcement learning processes, enabling them to call external tools like web searches and Python code execution. The company says that while the models are advanced, they currently lack the ability to generate images and audio, unlike their more powerful proprietary counterparts. The models come at a time when the AI landscape is becoming more competitive, with Chinese labs, including DeepSeek, and Meta's Superintelligence Lab developing open-source models. OpenAI's move is seen as a way to assert its leadership in the open AI sector, responding to both developer demands and growing pressure to support AI that aligns with democratic values. While OpenAI's new open models have achieved state-of-the-art performance on various benchmarks, the company has acknowledged challenges such as the models' tendency to 'hallucinate' more frequently than larger, more advanced models like GPT-4. The release comes amid broader global discussions about AI safety and ethical considerations, with OpenAI assuring users that the models were designed with safeguards in place. OpenAI has also opted not to release the training data used to develop these models, amid concerns over the potential misuse of copyrighted materials. Our open models are here. Both of — OpenAI (@OpenAI) August 5, 2025 The introduction of these models is expected to enhance research, development, and commercial opportunities, as enterprises can monetise them under the Apache 2.0 license without seeking further permissions from OpenAI.
