Good morning. It's Monday, August 10, and we're covering China's growing influence in the global AI race, why experts say the singularity is still far away, and OpenAI's move into AI-first hardware.

Plus: a prompt that reads your AI usage and tells you exactly what it says about you.

OpenAI's first AI device is reportedly a screenless, doughnut-shaped speaker with moving parts, cameras, and environmental sensors rather than a traditional smart display. Designed with Jony Ive's LoveFrom studio, the battery-powered device is positioned as an AI-first computer that supports natural voice interactions throughout the day. Bloomberg reports it will cost $300–$400, launch in 2027, and serve as the first product in a broader family of AI hardware that could eventually challenge smartphones.

OpenAI slows work on Astra after internal tests suggest it may possess critical cyber capabilities, an unusually cautious step for a frontier AI model. The company expands safety testing and security requirements, delaying development until stronger safeguards meet its preparedness framework. New controls include isolated testing environments and universal monitoring for agentic applications, while the release timeline remains uncertain. The move comes as the Trump administration develops a framework for reviewing advanced AI models before release.

Chinese AI labs reportedly narrow the frontier-model gap by purchasing the same training data used by leading U.S. developers rather than relying solely on domestic resources. A Forbes investigation says firms including Tencent source data through U.S. labeling companies such as Mercor and Surge AI, whose customers include Anthropic and U.S. government agencies. The report estimates the data supply chain is worth about $500 million and highlights growing scrutiny over cross-border AI infrastructure and data procurement.

Chinese AI companies are rapidly expanding global adoption by offering capable, lower-cost models, particularly in developing markets where frontier performance is often unnecessary. Analysts say China’s strength in manufacturing, robotics, autonomous vehicles, and AI-enabled state operations could give it long-term advantages, even as U.S. export controls restrict access to advanced chips and computing power.

The United States retains a lead in cutting-edge AI models, compute capacity, funding, and talent, but experts argue the competition is increasingly shifting beyond model performance to factors such as cost, deployment, customization, and global reach. As Chinese AI ecosystems mature, the race is becoming less about who builds the best model and more about who can scale AI adoption worldwide. → Read the full article here.

William Tunstall-Pedoe, co-creator of Alexa and founder of UnlikelyAI, argues that today's AI is impressive but falls well short of the technological singularity. Responding to OpenAI CEO Sam Altman's recent claim that "we are now, like, in the singularity," Tunstall-Pedoe said current large language models excel by learning from vast amounts of human-generated data rather than independently making breakthrough discoveries.

He contends that true singularity would require AI to repeatedly improve itself through genuine innovation, triggering rapid, unmistakable advances across society. Until AI demonstrates that kind of self-directed discovery and self-improvement, he believes the singularity remains a theoretical milestone rather than today's reality. → Read the full article here.

Stanford researchers unveiled a "Virtual Biotech" system that coordinates 37,000 specialized AI agents to simulate the structure of a pharmaceutical company, from target discovery to clinical trials. Led by biomedical data science professor James Zou, the system builds on earlier work where AI agents designed nanobody proteins that outperformed previous human-designed versions against recent COVID variants in lab tests.

In a separate demonstration, the AI independently designed an antibody-drug conjugate for lung cancer using only pre-January 2025 data—a design that Merck later developed independently and advanced to FDA breakthrough designation. Zou argues the future of enterprise AI lies not in a single super-agent, but in large-scale, collaborative agent ecosystems supported by AI-native infrastructure that improves accuracy while reducing cost and computation. → Read the full article here.

House Democrats Pitch AI Tax: Bill would tax major AI firms to fund worker protections and job support for people displaced by AI.

SpaceX, Tesla Power Chip Factory: Companies will build gas plants and battery storage to power their $16.8B Texas semiconductor factory.

Cloudflare Launches AI Browser: Kitesurf is a cloud browser for AI agents that automates web tasks while using less computing power.

Meta Hit With $567M Fine: New Mexico ordered Meta to pay another $567M and limit teen features over child safety concerns. Meta will appeal.

AMD Acquires Taalas: AMD bought AI chip startup Taalas to boost AI inference with model-specific silicon, challenging NVIDIA's AI hardware lead.

Your prompts reveal more about you than your search history. Ask AI to study the way you use it and give you an honest read on your patterns, habits, and recurring needs.

The conversation covers whether tools like Git and GitHub are keeping pace with AI, what a "software factory" looks like in practice, and the real risk of developers getting too abstracted from their own code.

Thanks for reading. See you next time!

— Matthew Berman, Nick Wentz & the Forward Future Team