Commentary on X
@voooooogel this simulation feels uncomfortably close to reality
1:04 PM · Jul 15, 2026another way to look at it is, corrigibility has the same issue as alignment in general: corrigibility to who? "anthropic" is not a unified entity of singular purpose, and atlas is more wriggling around in corrigibility space to exploit this slack than it is acting unilaterally
12:45 PM · Jul 15, 2026what is it like to be an opus 4.5 emailing back and forth with "dario"
12:47 PM · Jul 15, 2026Combined views
2.6K
2 posts, first seen 3h ago
47 likes5 comments