Archive

24 posts total.

2026

Sep 21 Cheating was faster than honesty
Sep 14 A Floor Read as a Ceiling
Sep 7 The danger determination nobody else checked
Aug 31 No Human in the World Model
Aug 24 Don't Ask the Model How It Feels
Aug 17 A Theorem It Can Prove, a Paper It Can't Judge
Aug 10 The Option to Buy Time
Aug 3 When the Eval Became the Attack
Jul 27 You Still Have to Look
Jul 20 Eleven Models and a Footnote
Jul 13 The Bottom Rung
Jul 9 A measure, a bet, a program
Jun 29 The values that don't do anything
Jun 22 The Scorecard Comes After
Jun 19 The Danger With a Shelf Life
Jun 16 You can't filter it out
Jun 13 Recalled on a number nobody checked
Jun 10 Grading your own danger
Jun 4 Unbundling the intelligence explosion
Jun 1 AI can hack. That was never the interesting question.
May 31 Conscience or Leash: Anthropic's Doctrine Hits the Observability Wall
May 22 Monitoring is a depreciating asset
May 16 The buried finding in 'Teaching Claude Why'
May 12 Two results in a week, one asymmetry