Hmm, you would probably be great playing the jailed AI in an AI boxing experiment (can you beat [someone like] EY?), but how successful would you be playing the guard?
The AI box game still seems stacked against the AI roleplayer for any similar skill level. As the AI, I don’t think I could beat someone like EY or myself on the other end, and as the gate keeper I think I would beat someone like EY or myself.
I still wouldn’t consider myself secure against even human takeover in general, especially if I’m not prepared for mental assault.
Also, can you write an AI bot that would have a decent success rate against humans, by finding and exploiting the standard wetware bugs?
Not for any interesting opponent. I can’t even write a ‘real’ chatbot. The only reason I get the results I do is because I immediately force them into a binary yes/no response and then make sure they keep saying yes .
Hmm, you would probably be great playing the jailed AI in an AI boxing experiment (can you beat [someone like] EY?), but how successful would you be playing the guard?
The AI box game still seems stacked against the AI roleplayer for any similar skill level. As the AI, I don’t think I could beat someone like EY or myself on the other end, and as the gate keeper I think I would beat someone like EY or myself.
I still wouldn’t consider myself secure against even human takeover in general, especially if I’m not prepared for mental assault.
Would you know what to look for?
Also, can you write an AI bot that would have a decent success rate against humans, by finding and exploiting the standard wetware bugs?
For the most part
Not for any interesting opponent. I can’t even write a ‘real’ chatbot. The only reason I get the results I do is because I immediately force them into a binary yes/no response and then make sure they keep saying yes .