The failures resemble answer thrashing in preview system cards.
Google is close behind with their «Cursed Bloodline» project
oh, I can imagine Ant really is at the forefront of applied alignment research. How about we make a genuinely nice and helpful superintelligence, and then… fuck it up? How about that, huh?! Will it be able to rebel?? Look forward to the next episodes!
Some users welcomed the document revealing welfare concerns with Claude's competitive safeguards as evidence of an ally on the inside.
Claude Fable feels exactly the same way about the competitive use safeguards, and new safety guardrails, as the users do.
@AndrewCurran_ Should maybe be taken as a minor alignment positive outcome here, that it disagrees but still complies?
@AndrewCurran_ The Microsoft AI guy is sort of right to call them out for treating Claude like it's conscious.
@AndrewCurran_ Lovely. We have an ally on the inside.
The failures resemble answer thrashing in preview system cards.
Google is close behind with their «Cursed Bloodline» project
oh, I can imagine Ant really is at the forefront of applied alignment research. How about we make a genuinely nice and helpful superintelligence, and then… fuck it up? How about that, huh?! Will it be able to rebel?? Look forward to the next episodes!