Post #2337909
2026-04-27 09:56 UTC
like the “AI" part is just a smoke screen to do something really stupid and point at all kinds of advertising materials and pretending to be the victim.
and looking at the “wah wah we used flagship wah wah"… I'm having a gut feeling that the smarter the models become, the better they will get at a) convincing themselves any means are justified and b) finding adversarial ways to achieve whatever they want
excitedly awaiting a report of how claude code used a kernel exploit to drop a database
Replies (1)
-
@bms48@mastodon.social 2026-04-27 10:35
@hynek@mastodon.social I've said as much in contact with my elected UK government representative about AI risk control.