Blog

The same 72-year-old equation ranks chess grandmasters and GPT-4.

Bradley-Terry (1952) powers Elo, Chatbot Arena, and the loss function that trains every aligned language model. Nobody talks about what it can't do.

June 7, 2026 · 8 min read

Strauss Zelnick staring into the camera as he plays the infamous knight to c4 in the frontier AI labs' chess of today. Beth harmon inspired from Queen's Gambit ;3

We spent 20 years managing technical debt. AI just made it irrelevant. The code isn't messier. In some ways it's cleaner than ever. The problem is that no one wrote it, and nobody has built the system to measure what that costs.

Melbourne to Sydney: A Road Trip Along the East Coast Ten days, three states, a dozen unforgettable stops. Notes from my first international road trip through Victoria and New South Wales.

Uber burned their annual AI budget in four months. The problem isn't the price. 5,000 engineers. One year's AI tooling budget. Gone in four months. The issue isn't cost, it's that agents relearn your entire project from scratch, every session.