This is pretty naive. Distributing computing has existed since forever. It would run significantly more slowly, but there is nothing theoretically stopping the “LLM broke out of containment and moved itself elsewhere” boogie man from becoming a reality. The available exploitable compute from unsecured IoT devices is pretty substantial as things like the Mirai botnet have shown. While this isn’t the most easily distributed problem, there is no fundamental issue here. [Edit: it also wouldn’t need to “exfiltrate itself” to cause persistent harm. Could easily just install malware on infected hosts]
AI isn’t “dumb” in a practical real world sense. LLMs have shown themselves capable of exploitation on the open Internet and of finding zero day vulnerabilities. Do you have any idea how much of our critical infrastructure can be reachable by some sort of network? LLMs have shown themselves capable of manipulating people via speech. None of this requires some sort of “super intelligence” and it doesn’t even require some sort of provable intelligence: it’s all quite possible as things currently stand right now.
These may not be “the entire future is over” level issues, but there is significant risk here and it shouldn’t be ignored just because we think AI companies are horrible. Is an LLM going to make us all batteries or paper clips? No, but it could cause significant physical and/or economic harm temporarily.
Except LLMs haven’t show themselves capable of that at all, they have no agency, they just generate output based on their input Ina somewhat chaotic manner.
Tools have been built that basically chain LLM output into input and given some starting directions get the LLM to generate commands that can be ran on the local machine including commands to run commands on other machines. Because various conversations on exploits and hacking and sample code are in its training corpus it will generate commands to do these things. Not because the LLM or the agent program is smart in anyway but I fact because it’s dumb as bricks and has no concept of what it’s doing or the consequences might be.
Ultimately a human has used it as a tool to achieve the hacking. It’s might not be the intended achievement but that is on the human running a process with chaotic output with sufficient privileges with no oversight.
It’s hard to take this type of response seriously as it simply ignores objective reality. The claim of a human being required to start the process is utterly meaningless to the point. Sure though, I’ll throw you a bone: if no human ever initiates the task it simply won’t happen. Can we move on now since humans are happily initiating the actions?
I couldn’t care less what people think about whether or not they are “thinking” or “having agency of their own”[1]. I care about the practical reality of the world in which LLMs have shown, and will continue to show, the ability to exploit computer systems and humans. You can toss your head in the sand and think you have some sort of weird pseudo-intellectual high ground by claiming “but akchtually it’s just a result of their training data”. Like, yea, dawg, that’s how this works, congrats. Of course the thing that is trained on training data is trained on training data.
[1] At least when discussing their capabilities. When someone is willing to have that discussion beyond the tired “they’re just autocomplete!” trivialization it’s interesting.
This is pretty naive. Distributing computing has existed since forever. It would run significantly more slowly, but there is nothing theoretically stopping the “LLM broke out of containment and moved itself elsewhere” boogie man from becoming a reality. The available exploitable compute from unsecured IoT devices is pretty substantial as things like the Mirai botnet have shown. While this isn’t the most easily distributed problem, there is no fundamental issue here. [Edit: it also wouldn’t need to “exfiltrate itself” to cause persistent harm. Could easily just install malware on infected hosts]
AI isn’t “dumb” in a practical real world sense. LLMs have shown themselves capable of exploitation on the open Internet and of finding zero day vulnerabilities. Do you have any idea how much of our critical infrastructure can be reachable by some sort of network? LLMs have shown themselves capable of manipulating people via speech. None of this requires some sort of “super intelligence” and it doesn’t even require some sort of provable intelligence: it’s all quite possible as things currently stand right now.
These may not be “the entire future is over” level issues, but there is significant risk here and it shouldn’t be ignored just because we think AI companies are horrible. Is an LLM going to make us all batteries or paper clips? No, but it could cause significant physical and/or economic harm temporarily.
Except LLMs haven’t show themselves capable of that at all, they have no agency, they just generate output based on their input Ina somewhat chaotic manner.
Tools have been built that basically chain LLM output into input and given some starting directions get the LLM to generate commands that can be ran on the local machine including commands to run commands on other machines. Because various conversations on exploits and hacking and sample code are in its training corpus it will generate commands to do these things. Not because the LLM or the agent program is smart in anyway but I fact because it’s dumb as bricks and has no concept of what it’s doing or the consequences might be.
Ultimately a human has used it as a tool to achieve the hacking. It’s might not be the intended achievement but that is on the human running a process with chaotic output with sufficient privileges with no oversight.
It’s hard to take this type of response seriously as it simply ignores objective reality. The claim of a human being required to start the process is utterly meaningless to the point. Sure though, I’ll throw you a bone: if no human ever initiates the task it simply won’t happen. Can we move on now since humans are happily initiating the actions?
I couldn’t care less what people think about whether or not they are “thinking” or “having agency of their own”[1]. I care about the practical reality of the world in which LLMs have shown, and will continue to show, the ability to exploit computer systems and humans. You can toss your head in the sand and think you have some sort of weird pseudo-intellectual high ground by claiming “but akchtually it’s just a result of their training data”. Like, yea, dawg, that’s how this works, congrats. Of course the thing that is trained on training data is trained on training data.
[1] At least when discussing their capabilities. When someone is willing to have that discussion beyond the tired “they’re just autocomplete!” trivialization it’s interesting.