Writing
Notes
Mostly about AI reliability: when a system should say it doesn't know, how to check that an answer is backed by something real, and how to reason about the bad end of a distribution.
- Let it say I don't knowJun 18, 2026
A system that always gives you an answer is easy to build and hard to trust.
- The worst likely dayMay 2, 2026
A go/no-go decision is a bet on the tail, not on the average. Most risk scores model the average.
- Against people, or for themMar 11, 2026
I learned to build by taking games apart. The useful part was not the exploits.