Conversation

Do not publish capabilities research. There is sometimes a case for doing it, there is only far more rarely a valid justification for publishing it.
Quote Tweet
I think it's generally bad to work on AI capabilities. Not always (since alignment tech is often dual-use), and not forever (since never building TAI seems awful). But broadly, I think we should not build extremely powerful AI systems before we figure out how to make them safe.
If they aren't publishing, the research community doesn't have visibility into their current work; which means no chance to raise concerns about the directions that work may be taking, or the safety protocols used. That doesn't seem like the an optimal situation.
1
Show replies