I spent some time digging through the METR time horizon paper. I thought I'd hate it. It's actually a lot more rigorous than many things in this space (e.g. AI 2027). Credit where credit is due. It's pretty good.
There's been a lot of recent concern about existential risk from power-seeking AI. I push back against that concern in a new paper: https://t.co/DHhtlbr3RT
I wish that important concerns about AI misbehavior were not so often mixed with speculative concerns about existential risk. These letters would be much easier to sign if they weren't mixed. https://t.co/7RpPIVfT9F