Blog
Keeping Things on Track: How PID Controllers Work
Three ideas, three terms, a formula you could write in a single line. How PID controllers solve one of the most universal problems in engineering.
Kelly in the Wild: Against the Ensemble
How Kelly sizing, LMSR pricing and real-time Bayesian updating fit together into a working prediction market architecture.
The Clever Trick at the Heart of RLHF
On why ELO — a 1960s chess rating system — sits at the core of how modern language models are trained.
ICML 2025 Review
Notes and insights from ICML 2025, covering cutting-edge research in robotics and reinforcement learning.