It’s a pretty gestalt feeling. I’ve had the feeling for a while, and it’s been stronger on recent model update posts. Some sense of missing mood, that the concerns about risk-to-world from new models are in some sense missing or suppressed. That he isn’t avoidant of the incremental version of the bad thing that shows up in his broad-topic update posts as “aligning a smarter-than-human AI is hard”. Or that he’s perhaps getting punked by deceptive alignment on the part of the orgs and/or models. That he isn’t averse to including models showing off/peacocking as though it’s fun/mildly good. But I’m not sure about any of these; what I’m more sure about is that the taste of metaphorical verbal sugar alcohols is stronger in his recent posts. (I did read the entire post before commenting.)
It’s a pretty gestalt feeling. I’ve had the feeling for a while, and it’s been stronger on recent model update posts. Some sense of missing mood, that the concerns about risk-to-world from new models are in some sense missing or suppressed. That he isn’t avoidant of the incremental version of the bad thing that shows up in his broad-topic update posts as “aligning a smarter-than-human AI is hard”. Or that he’s perhaps getting punked by deceptive alignment on the part of the orgs and/or models. That he isn’t averse to including models showing off/peacocking as though it’s fun/mildly good. But I’m not sure about any of these; what I’m more sure about is that the taste of metaphorical verbal sugar alcohols is stronger in his recent posts. (I did read the entire post before commenting.)