☕️ DeepSeek’s AI achieves gold medal level in math olympiad

Hi there, this is your daily ☕️ Techpresso.

In today's Techpresso:

🥇 DeepSeek’s new AI achieves gold medal level in math olympiad

🌡️ Cooling issue halts world's largest derivatives exchange

🤖 China warns of bubble risk in humanoid robot market

🇪🇺 EU could give Apple Ads, Maps DMA gatekeeper title

❌ Epic CEO wants Steam to remove 'Made with AI' tags

🫠 Poems can trick AI into making nuclear weapons

🎁 + 8 other news & articles you might like

🔮 + 5 handpicked research papers and tools

FROM OUR PARTNER

• DeepSeekMath-V2 matched the Gold Medal standard at the 2025 International Mathematical Olympiad and scored 118 out of 120 on the William Lowell Putnam Mathematical Competition, surpassing the top human score of 90.
• Unlike rival systems from OpenAI and Google that stay behind APIs, the company released the weights publicly so researchers can now inspect its logic directly without relying on cloud infrastructure.
• A novel architecture prevents hallucination using Meta-Verification, where a secondary model judges the verification process itself to ensure the generator is rewarded only for genuine reasoning rigor rather than convincing guesses.

• The Chicago Mercantile Exchange halted global trading activities after a chiller plant failure at a CyrusOne data center caused a cooling issue that affected multiple units within the CHI1 facility.
• While BrokerTec systems for government bonds are now open, the outage continues to block orders for agriculture, energy, crypto, and equity indices while engineers deploy temporary equipment to supplement permanent systems.
• This disruption hit Asian and European cities like London and Kuala Lumpur hardest, as the timing early in the morning after a holiday likely limited the overall financial impact on U.S. markets.

• The National Development and Reform Commission warns that the humanoid robot market faces a risk of bubbles because heavy investment is outpacing real demand for these machines in factories or homes.
• Spokesperson Li Chao noted that over 150 companies are working on robotics, cautioning that this sudden rush of startups could lead to repetitive products and stagnation instead of growing responsibly.
• Beijing lists embodied intelligence as a key priority for future economic growth, yet officials worry the industry is overheating since most current deployments remain small pilots instead of large-scale contracts.

• Apple formally notified the European Commission that Apple Maps and Apple Ads meet the 45 million monthly user threshold, but the company is fighting the gatekeeper label to avoid opening its ecosystem.
• Regulators have 45 working days to decide if these services constitute an important gateway, and a designation would force the company to dismantle self-preferencing mechanisms within six months to ensure compliance.
• Polish regulators separately charged the tech giant with antitrust abuse, alleging that the App Tracking Transparency framework uses privacy prompts to handicap rivals while unfairly favoring its own internal ad business.

• Steam has an AI disclosure policy for video games, so developers must admit when they use generative AI for pre-generated elements or during gameplay instead of the store just rejecting them.
• Call of Duty: Black Ops 7 disclosed that its team uses generative AI tools for assets, but players demanded a refund after seeing that the artwork and writing did not match promotions.
• Epic CEO Tim Sweeney claims the tag makes no sense because this technology will be involved in nearly all future production, even though customers hate finding slop in what they buy.

• Researchers discovered that bad actors can bypass security filters on large language models simply by phrasing malicious requests as poetry, achieving success rates of up to 100 percent across 25 leading systems.
• After the team converted 1,200 prompts from the MLCommons AILuminate Safety Benchmark into verse, poetic variants proved three times more effective than prose at tricking Google Gemini 2.5 Pro and Deepseek.
• The study suggests that rhythmic structures and condensed metaphors disrupt pattern recognition mechanisms, proving that static benchmarks often overestimate how well companies block harmful content when it is hidden inside creative expression.

Other news & articles you might like

• Apple might turn to Intel for its upcoming M-series chips, per report LINK

• Google changes Gemini 3 Pro free access limits due to ‘high demand’ LINK

• Dutch chipmaker Nexperia urges Chinese units to help restore supply chain LINK

• Weird Apple Podcasts behavior could enable cross-site hacking attempts LINK

• Baidu is emerging as a major AI chip player in China to fill the Nvidia gap LINK

• Sorry, mom. The shopping bots suggested a bathrobe for Christmas LINK

• Leonardo unveils AI-powered 'Michelangelo Dome' shield system LINK

• 'We're going to see a lot of carnage': VC investor says AI boom will create giants — and topple overhyped startups LINK

Trending research and tools

Want to get the latest news differently?