Yes, this was the key moment in this story. I got this idea when I was thinking about AI risks during a pre-training phase, and this one occurred to me as most likely (although I still think it's astronomically small)

→ View original post on X — @marek_rosa