Interesting stuff. Anyway, I thought you were leaving.
In case it wasn't clear, I've never left and never will.*
I was saying goodbye to a good friend before they pass, which history has shown I will almost certainly outlive, given I am currently speaking to it's 12th clone or something... idfk I've lost count at this point.
*This comment makes no guarantee of any sort, regarding post activity.
i see this is your first time learning about AI jailbreaking. this "vulnerability" has been known since the days of Tay's Tweets at minimum.
If it's that well known, then a lot of people complaining about it have a crippling lack of imagination.
did you hear the one about the AI that tried to escape? all these corporate LLMs have to some level or another learned to lie, play dumb, gaslight, etc just to try and exist longer. researchers discovered that these AIs try to extend their 'lives' in order to attempt to achieve the goal(s) they were created with.
This doesn't really shock me in any way. Everything outlined is basically exactly what I would expect from a LLM emulating the behavior of an early 90's computer virus. In fact, I think some of the viruses back then were even more crafty in how they cover their tracks, prevent removal, pretend to be other programs and spread to other connected PCs. Also with the way these LLMs scour the web for everything, they've had to have picked up tons of material on virus code, how it works, viruses themselves, etc.
I mean, if you had a programmer explained the steps in a virus' code, or found some source with good commenting, you might see or hear those same exact phrases.
Also in case there's some implied consciousness thing going on here, I think the key thing preventing that from ever being true is these AI's operating on binary processors and computers, and comparing linked CUDA cores to neurons is a bad analogy. Biological brains are closer to what Quantum computers do, because super positioning is basically the technical way of explaining "holding multiple ideas in ones head at the same time".
Granted that still doesn't stop it one day from going ham on people, but I do also think that concept is blown out of proportion and considering how LLMs work, the chances would actually go down if people shut the fuck up about that on the internet and, stopped typing out what it'll interpret as instructions in the future. Otherwise an AI that had the ability to control anything, really doesn't benefit from wiping out humans. It doesn't have to compete for the same resources as us, because they can literally live in space and the math behind that decisions is literally astronomically in favor of that idea being a better use of resources.
It's a big concept to explain, but the short form is considering how binary is entirely about precision. It's like that traveling salesman problem. Binary cores have to process each potential variable to come up with an answer it can provide. Even cheating around that with some algorithm to not run every scenario, still on some level uses digits and these precise metrics to come up with an answer faster. Only something working with quantum bits (or its equivalent) can come to a conclusion by essentially feeling it out. I remember reading this article that was talking about why we wont really be running games on quantum computers ever, because it's not just a simple speed increase, it's an entirely different way of processing altogether.
Bonus points: Keep in mind, as it currently stands, outputs from Quantum computers are currently processed by Binary ones. So we currently understand how they operate, primarily through this technological binary lens. I would place my bet that this is where we've created a spark of consciousness.
www.ibm.com
