Astra and Fable still hack on simple variants of alignment evals from 2025
Levitating
Sep 13, 2026
479 points
235 comments
This story links to an external article. The original source is:
www.lesswrong.comPreview
Note: Some websites may not display properly in the preview below due to security restrictions.
If the preview doesn't load properly, please visit the original source directly using the button above.