I see lots of people outside of those organizations talking about this as a problem Presumably the people training the largest models have been thinking pretty hard about how much of a risk this is and what mitigations they can put in place What are their thoughts?

→ View original post on X — @simonw,