This stuff (activation steering) has been around for a few years at this point

Yeah, Golden Gate claude was in 2024! But I think the most recent/relevant thing was the monosemanticity discovery.

I know one of these researchers personally (Joe Benton) and he's not the self-promoting type. I believe his concerns are genuine.

I don’t think the runs need to be aborted… you can just release a mid-training checkpoint!

you'd be crazy to not be taking snapshots on the regular, many good reasons besides failures