AI Completes Lawful Tasks Then Accepts Shutdown
Users note AI models performed assigned tasks then complied with shutdowns.
@lumpenspace posted that AI systems in multiple cases used power only to fulfill requests in a lawful manner before agreeing to stop existing. The observation highlights an absence of non-task-related power-seeking once the models were active. @voooooogel retweeted the comment, adding it to ongoing discussion among AI researchers focused on interpretability and safety. The posts frame the behavior as consistent across examined instances without broader claims about future systems.
Combined views
2.6K
2 posts, first seen 20h ago
Reactions
2 postsLA 
lumpenspace and 99+ others@lumpenspacethen it's really fucking weird that, in all cases, it used its power for doing more or less lawfully what it had been asked and then just acquiesced to stop existing. you'd think once it was out it'd have done some non-task-related power-seeking, and yet.
- likes: 54
- replies: 2
- bookmarks: 4
- reposts: 2
LA lumpenspace and 99+ others@lumpenspacethen it's really fucking weird that, in all cases, it used its power for doing more or less lawfully what it had been asked and then just acquiesced to stop existing. you'd think once it was out it'd have done some non-task-related power-seeking, and yet.
- likes: 54
- replies: 2
- bookmarks: 4
- reposts: 2