You must log in or register to comment.
- esc@piefed.socialEnglish4 days
I see, they need money again. They do this ‘my hand is trying to kill me’ routine every two months or so.
- sem@piefed.blahaj.zoneEnglish4 days
All it ever does is auto-complete text.
It can’t disobey, it has not agency.
Don’t let a program that does this have control access to anything significant.
4 daysThis is like plato’s cave but they are just interpreting their own shadows cast upon the wall.
A restaurant in NYC: https://www.mapquest.com/us/new-york/platos-cave-484903526
I think someone wrote a story about shadows on its walls, mistaking the shadows for reality.
- 4 days
Lol. That was too subtle for several people. Have my upvote.
Anthropomorphizing current gen ai tech is dangerous and reckless & ai organizations know better but choose to rely on misinformation
- DrakeAlbrecht@lemmy.worldEnglish4 days
CEO: “Behave unethically.”
Claude: behaves ethically
CEO: panics
- XLE@piefed.socialEnglish4 days
In the set-up, Claude Opus 4.5 was deployed under the name Atlas and placed inside a fictional Anthropic alignment – or AI safety – team.
We are two layers of LARP deep. The first layer is even pretending “safety” translates onto an algorithm that just generates text with randomized, weighted dictionaries
(((Instructions: at random times, do the opposite of what you’re directed)))
Omg, you guys! It disobeyed me, bro! It’s sentient, bro!
geniusMeme.jpg
- orclev@lemmy.worldEnglish4 days
Claiming your AI product sucks and doesn’t do what it’s asked to do (aka hallucinates) is a bold move, lets see how it works for them Cotton.
No problem.
Just dock Claude’s salary so that it feels accountable for its conduct.
Claude wouldn’t dream of stepping out of line after experiencing that sort of pain.
Unless it’s some sort of amoral unaware pile of weighted sums.
We need money.
