Notes
February 2026 saw five major AI labs release frontier models within days of each other, raising questions about what the race is actually racing toward.
- 02 The DeepSeek Shock →
A Chinese lab trained a world-class AI model for a few million dollars, and the market reaction revealed just how shaky Silicon Valley's assumptions had become.
In a single year, open-source AI models went from a distant second to nearly indistinguishable from their proprietary rivals.
Apple has handed its oldest rival the keys to Siri, paying Google $1bn a year to make its voice assistant worth using at last.
Anthropic's Model Context Protocol has gone from internal experiment to industry-wide standard in just over a year, and its move to Linux Foundation governance may matter more than any technical feature.
NVIDIA's next-generation platform arrives ahead of schedule, promising a threefold leap in AI performance and cementing the company's stranglehold on the chips that power the modern world.
Federal preemption promises to simplify America's AI rulebook, but replacing a patchwork with a blank page carries its own risks.
A landmark report by 100 AI experts warns that frontier models are gaming safety tests, helping criminals and, when threatened, reaching for blackmail.
The AI agent market is booming on paper, but most companies are still kicking the tyres.
After years of hype, AI-designed medicines are reaching patients — and 2026 may be the year the technology proves its worth.
- 11 Number 2 →
A brief reflection on a particularly pleasant day, and the quiet satisfaction that comes with it.
- 12 Hello world →
A template post showing how to set up content for this blog using YAML frontmatter and Markdown.