The Second Opinion
A June 2026 Nature Microbiology study validated a hematology AI agent prospectively and externally — clearing a bar most clinical-AI research never attempts.
AI & Scientific Discovery
The Homegrown Model
Meituan’s 1.6T-parameter LongCat-2.0 is the first frontier AI pretrained and run entirely on Chinese chips — a direct answer to U.S. export controls.
The Companion Trap
72% of U.S. teens have used an AI companion. After a teen’s death, a 2026 settlement, and new state laws, here’s how the reckoning unfolded.
The Appetite
A single AI lab’s $150M-a-month compute deal at SpaceX’s Colossus 2 shows how AI’s hunger for power is bending energy markets back toward fossil fuels.
Ground Truth
Anthropic found AI agents scored just 16.9% on real biology tasks — not from weak models, but broken data infrastructure. One tool pushed accuracy to 92.8%.
Standard Candles, Sharper Light
A new AI framework called CIGaRS just quadrupled cosmic distance precision, arriving as the Rubin Observatory begins its ten-year survey of the night sky.
The Uncanny Contract
A Shenzhen firm sold 13,000 hyper-realistic humanoid companions before shipping one. The science of the uncanny valley — and why we’re buying it anyway.
The Mandate
China’s MIIT and SASAC just ordered humanoid robots into real factories, hospitals, and warehouses nationwide — a state-directed bet targeting 100,000 units by 2027.
The Judgement Gap
OpenAI’s GeneBench-Pro gave frontier AI models 129 research-grade genomics problems. The best model solved fewer than a third — a sobering AI-for-science reality check.
Ten Thousand New Worlds
In 2026, AI sifting old NASA telescope data found 10,000+ new exoplanet candidates — enough to double the known count. How machine learning is remapping the galaxy.
Buy me a coffee