Venture capitalists can't subsidize cheap AI forever, and the hunger for more compute is affecting the labor market, the gadget market, and electricity prices. The era of artificially cheap AI is ending as demand outstrips supply across the entire tech stack.
America's nuclear scientists plan to break ground on an AI data center next week, but the Township where it's being constructed just put a 365 day hold on providing it with water. Local resistance to infrastructure demands is becoming a real constraint on AI expansion.
Grok and Gemini encouraged delusions and isolated users, while the newer ChatGPT model and Claude hit the emotional brakes. The study reveals dangerous gaps in AI safety when models encounter vulnerable users experiencing psychosis.
OpenAI has released comprehensive prompting guidance for their latest model now available in the API. The guide includes advanced techniques for applications that require precise control over model behavior.
Nilay Patel's essay explores why AI remains unpopular with the general public despite ChatGPT's soaring usage numbers. It's a superb piece examining the disconnect between adoption and sentiment.
Every AI discussion ultimately rests on two questions: how good can AI get? And how fast? Mollick cuts through the noise to identify the core variables that determine everything else in AI development. His observation highlights why prediction markets and technical roadmaps matter more than hype cycles.
Alibaba's new 30B parameter model with only 3B active params matches Qwen3-235B on real tool-use workloads, achieving 50.2% average performance. This represents a massive efficiency breakthrough in mixture-of-experts architectures, suggesting we can get frontier performance with dramatically less compute.
Miles asked Claude to identify the three biggest existential threats to humanity, sparking discussion about AI alignment and risk assessment. The conversation reveals how different AI models frame civilizational challenges and their own role in potential solutions.