What I’ve Learned and Built, and What I’m Learning and Building
-
Reliable Systems from Unreliable Models
Short essays on the reliability of AI systems. The series starts with von Neumann's 1952 lectures, which showed that redundancy can make a machine as reliable as you like, provided its parts fail independently, and then asks what changes when errors are correlated, as they are with language models.
-
A Primer on Price Optimization
Four posts that build up the math of price optimization one layer at a time: a single item, many items at once, the business constraints that couple them, and finally items that compete for the same demand. Full derivations, figures computed from the models themselves, and notes on what breaks in production.
-
A Primer on Ad Bidding
Three short posts that build up the math of bidding in ad auctions from the advertiser's side: a single auction, a day of auctions sharing one budget, and pacing that budget in real time. Derivations under the simplest assumptions, figures computed from the models themselves, and what each result means in practice.