Systems Market Human
AI Cracks an 80-Year-Old Math Problem — and Fails Basic Election Questions
OpenAI has disproved an Erdős conjecture that mathematicians had pursued for 80 years. Paul Erdős was among the most prolific mathematicians in history, and his conjectures are considered among the field's hardest problems. The achievement represents, researchers argue, a genuine intellectual contribution rather than computational brute force — a step toward systems capable of producing original knowledge rather than merely synthesizing existing information.
Jeff Bezos publicly described Amazon's secretive Prometheus project for the first time, calling it an 'artificial general engineer' — a system designed to navigate the simultaneous demands of physical constraints, material properties, safety requirements, and cost optimization that characterize complex engineering work. The framing positions Prometheus as a direct competitor for tasks currently performed by human engineers.
In the commercial AI landscape, Google's Gemini has overtaken all other ChatGPT competitors combined as an AI referral source, according to BrightEdge data. Gemini's web referral share nearly tripled in the first quarter of 2026 while ChatGPT contracted for the first time — a shift in actual usage patterns rather than media coverage. Nvidia posted record quarterly revenue of $81.6 billion, with CEO Jensen Huang citing accelerating global AI infrastructure buildout. Huang also acknowledged, however, that Nvidia has 'largely conceded' China's AI chip market to Huawei, calling the Chinese company 'very, very strong' — a remarkable admission of market loss even amid record global growth.
Senator Sanders warned that Meta is replacing workers with AI systems as new layoffs begin, exemplifying a broader transition from AI tools that augment human productivity to systems that substitute for it directly. Against that backdrop, a study found that AI chatbots fail to answer election-related questions accurately approximately 90% of the time — a striking limitation in systems otherwise achieving superhuman mathematical performance. The uneven capability profile suggests AI development is advancing in spikes rather than uniformly, with profound implications for how and where these tools are deployed.