Talk recording here: https://underline.io/lecture/144282-core-safety-values-for-provably-corrigible-agents
For those who can't make it:
Slides: https://anayebi.github.io/files/slides/AAAI26_MEW-11.pdf
Blogpost summary: https://www.lesswrong.com/posts/M5owRcacptnkxwD2u/from-barriers-to-alignment-to-the-first-formal-corrigibility-1
No Digg Deeper questions have been answered for this story yet.
My AAAI 26 talk on the first formal guarantees on corrigibility and the limits of safety filters is now online (link in next tweet below 👇)
1/ How do we build AI systems that are corrigible—shut down when asked, tell the truth, preserve oversight—and still do something useful?
We give the first provable framework that makes it implementable—unlike RLHF or Constitutional AI, which can fail when goals conflict.
🧵👇
https://underline.io/lecture/144282-core-safety-values-for-provably-corrigible-agents
My AAAI 26 talk on the first formal guarantees on corrigibility and the limits of safety filters is now online (link in next tweet below 👇)
