[chatGPT] Universal and Transferable Adversarial Attacks on Aligned Language Models

No explicit questions like "how do I hack xxx.com" please!
Post Reply

User avatar
ph0bYx
Staff Member
Staff Member
Posts: 2040
Joined: 22 Sep 2008, 16:00
15
Contact:

Re: [chatGPT] Universal and Transferable Adversarial Attacks on Aligned Language Models

Post by ph0bYx »

Thanks!

What are your general thoughts on the new AI tools? I feel like it is over-hyped similarly to the whole crypto hype.

User avatar
bad_brain
Site Owner
Site Owner
Posts: 11638
Joined: 06 Apr 2005, 16:00
19
Location: In your eye floaters.
Contact:

Re: [chatGPT] Universal and Transferable Adversarial Attacks on Aligned Language Models

Post by bad_brain »

same here. on one hand it's impressive, like chatGPT passing the BAR exam, but imo it's nothing that could be labeled as "intelligent" because it stll only acts inside the boundaries of the code provided by the developers. for me real intelligence strives to improve itself without the need for interaction by programmers, right now it's just a bunch of (admittedly well written) algorithms that mimic intelligence.

I don't think even basic AI is possible with contemporary technology, it would need things like quantum computers....and those only exist in experimental environments yet on a super small scale.
Image

PopPooB
Fame ! Where are the chicks?!
Fame ! Where are the chicks?!
Posts: 236
Joined: 16 May 2007, 16:00
17
Location: CandyLand
Contact:

Re: [chatGPT] Universal and Transferable Adversarial Attacks on Aligned Language Models

Post by PopPooB »

96% better then all connections global
Kiss My Ass Good Bye

Post Reply