Short version: you were significantly closer to the correct position than I was.
However, the position that would have been more correct than either of us at the time is “Wait, what do we even mean by alignment research and capabilities research? Can we even draw a clear line between them in order to figure out who to ostracize? If they’re so similar, then how can we justify thinking that one is extremely good and the other extremely bad?”
Such thinking could have led us to become much less confused about weird dynamics in alignment. I started doing that thinking in this post (from 2024), and continued it in this series. Better late than never!
4 years later: thoughts?
Short version: you were significantly closer to the correct position than I was.
However, the position that would have been more correct than either of us at the time is “Wait, what do we even mean by alignment research and capabilities research? Can we even draw a clear line between them in order to figure out who to ostracize? If they’re so similar, then how can we justify thinking that one is extremely good and the other extremely bad?”
Such thinking could have led us to become much less confused about weird dynamics in alignment. I started doing that thinking in this post (from 2024), and continued it in this series. Better late than never!