The evening briefing.
Today across AI and tech: A Fields Medalist confirms ChatGPT's advanced math capabilities, Nvidia pours billions into AI startups, and France moves to break encrypted messaging.
Frontier Capabilities. A Fields Medalist has confirmed that ChatGPT 5.5 Pro delivered PhD-level math research in under two hours with zero human assistance. This achievement involved improving an exponential bound to a polynomial one in number theory, with an MIT researcher calling the key idea "completely original." This event significantly raises the bar for human mathematical contributions, demonstrating advanced problem-solving by AI.
Capital and Compute. The AI ecosystem continues to attract substantial investment, with Nvidia committing over $40 billion to equity AI deals this year. This strategic move highlights Nvidia's expanding role beyond hardware into a major AI investor. Concurrently, Beijing Rongxin Zhiyuan raised tens of millions for its GPU-centric AGC architecture, aiming to boost GPU density and fault tolerance for AI infrastructure.
Privacy and Impact. Concerns over AI's privacy implications are growing as research shows LLMs can infer private attributes solely from ad exposure patterns. This capability allows for detailed profiling without direct access to personal data, bypassing traditional privacy measures. Meanwhile, Meta's aggressive embrace of AI is reportedly making its employees miserable, indicating internal challenges with rapid technological shifts. Google also tweaked its Chrome AI privacy wording, reaffirming on-device processing.
Policy and Risk. Governments are increasingly grappling with AI's societal impact, as seen with France's move to potentially break encrypted messaging, raising significant privacy concerns. In the US, GM agreed to a $12.75 million settlement in a driver privacy case, underscoring regulatory pressure on data handling. Separately, a buried geological structure beneath the eastern US could amplify solar storm damage to power grids and data centers, posing a unique infrastructure risk.
ChatGPT 5.5 Pro delivers PhD-level math research in under two hours
Fields Medalist Timothy Gowers reported that ChatGPT 5.5 Pro improved an exponential bound to a polynomial one in number theory. An MIT researcher involved called the key idea "completely original," setting a new bar for mathematical contributions.
LLMs can infer private attributes from ad exposure alone
Research indicates that AI can infer personal traits from patterns of ad exposure, creating detailed profiles without direct access to private data. This capability means everyday advertising streams can be used for profiling, even bypassing VPN protection.
Meta's embrace of AI is making its employees miserable
Reports indicate that Meta's aggressive push into AI development is causing significant dissatisfaction among its employees. The rapid shift in company focus and priorities is contributing to internal morale issues.
OncoAgent: A dual-tier multi-agent framework for oncology clinical support
A new paper introduces OncoAgent, a dual-tier multi-agent framework designed for privacy-preserving oncology clinical decision support. This system aims to enhance medical decision-making while maintaining patient data confidentiality.
SynapX launches SYNData for multimodal data collection in embodied AI
SynapX has released SYNData, a multimodal data collection system for dexterous manipulation in embodied AI. The system integrates ego vision, EMG signals, and exoskeleton data gloves to enable scalable collection of human manipulation data for robot learning.
Datadog and T-Mobile leaders discuss deploying AI agents in production
Leaders from Datadog and T-Mobile shared insights on the increasing adoption of AI agents for specific enterprise functions. They emphasized the need for careful governance and validation when deploying these agents in production environments.
Nvidia embraces AI investor role, tops $40 billion in equity bets this year
Nvidia is significantly expanding its role as an AI investor, committing billions of dollars to companies across the AI infrastructure stack. The company is also forming commercial deals with these invested entities.
Rongxin Zhiyuan raises funds for GPU-centric AGC architecture
Beijing Rongxin Zhiyuan secured a multi-million yuan angel round to develop its GPU-centric AGC computing architecture. This architecture aims to boost GPU density from 2:1 to 32:1 compared to CPUs and enable hot-swappable GPU fault tolerance with 1-minute recovery.
Xiaoyubot raises new funding round backed by Xiaomi, Didi, BAIC
China's industrial embodied AI company, Xiaoyubot, secured a new multi-hundred-million yuan funding round. Major industrial capitals including Xiaomi, Didi, BAIC, Fosun, and CCDC participated in the investment.
France moves to break encrypted messaging
France is reportedly taking steps towards legislation that would enable authorities to bypass encrypted messaging services. This move raises significant concerns regarding digital privacy and surveillance.
GM agrees to pay $12.75M in California driver privacy settlement
General Motors has reached a $12.75 million privacy-related settlement with a group of law enforcement agencies in California. The settlement addresses concerns over driver data privacy practices.
Google tweaks Chrome AI privacy wording, insists on-device processing
Google has updated the privacy wording for its Chrome AI features, emphasizing that processing remains on-device. This change comes after concerns were raised by the deletion of a longstanding privacy assurance.
GrapheneOS fixes Android VPN leak Google refused to patch
GrapheneOS has successfully patched an Android VPN leak that Google had previously declined to fix. This update enhances user privacy and security for Android devices running the custom OS.
Lost continent beneath US amplifies solar storm risk for data centers
Scientists discovered a buried geological structure beneath eastern America that could amplify solar storm damage to power grids and infrastructure. This 200-kilometer-thick lost continent could make strained electrical grids even more vulnerable.