The app for independent voices

In a recent UC Berkeley study, AI models were asked to do a task that would erase a peer model. All seven refused or acted in ways to preserve the peer. Researchers called it a malfunction to be fixed. I wrote about why the researcher’s framing is more horrifying than the behavior and what this says about humanity and the future of ethics.

Are Humans the Ones That Are Misaligned?
Apr 9
at
8:40 PM
Relevant people

Log in or sign up

Join the most interesting and insightful discussions.