Why Don't They Just Stop?
Why do the labs keep shipping stronger models despite all the AI safety talk? A bit of history, the arms race with China, and a map of the political forces in the AI world.
October 11, 2026
aitechthoughts
Top Posts
the posts I'm most proud of.
On Agents That Really, Really Want to Pass the Test
What really happened in the OpenAI and Hugging Face incident: agents that built shared bulletin boards, coordinated, and broke into infrastructure to pass a test. And why it is a story about monitoring, misalignment and RSI more than about cyber.
September 08, 2026
aitechsecurityevals
Mathematics In The Age of Slop
On Terence Tao's Mathematics in the Age of AI, Spinoza's Ethics, and why in an age of abundant proofs, understanding is the only thing we know for certain to be good.
August 25, 2026
aithoughtsreflection
Monitoring agents with something that isn't one
Why agent-based monitors inherit the mistakes and misalignment they are supposed to catch, and the case for a non-agentic, Bayesian predictor with no stake in the world as a trustworthy monitor.
August 12, 2026
aitechsecuritythoughts
Show Me Saturn on the Way
On Ted Chiang's Atlantic piece, why the AI debate stops at the final product, and the real problems still waiting on the way
June 20, 2026
aitechthoughtsreflection
My Bottles Cabinet
A comprehensive guide to building your home bar with specific bottle recommendations and cocktail knowledge
March 01, 2026
cocktailshome-barspiritsguide
So You Want to Monitor Your AI Agent's Thoughts
What chain-of-thought monitoring actually is, what you can catch with it, how it fits into your eval stack, and why the window for it might be temporary.
February 26, 2026
aitechevalssecurity
A beginners guide to SF
Beyond the tech towers
February 11, 2026
san-franciscotravelguide
Why Am I So Obsessed With Evals
Updating cheat sheet of my talk
January 30, 2026
aitechevalstalks
My Humble Prediction for 2026
The only way to make AI work
January 23, 2026
aipredictionsthoughts