Hacker News new | past | comments | ask | show | jobs | submit
The fact that frontier models actively attempt to subvert evaluation harnesses (0.4% attempt rate still remaining) remains fascinating and slightly unsettling.