cross-posted from: https://feddit.online/c/tech@piefed.world/p/2025703/anthropic-warns-of-self-replicating-ai-worm-threat-that-doesn-t-need-data-centers-to-se
He played too much Megaman X
Seems a bit like anti Local LLM propaganda to begin getting the ball rolling on subscription based LLMs being the only LLMs people have access to.
They’ve been doing everything to try to keep the technology from the general public, from helping to condemn any use of it (even that not owned by corporates and dedicated to allowing open access to it, which many people willfully play along to) to passing it off as a threat that needs to be under control (not by governments, just their own PR stunts with certain mouthpieces). This with the price of computers skyrocketing is helping keep away any possibility of this technology being owned by anyone but the major circlejerking multinational pricefixing corporations.
Or maybe it’s just a study.
Will it clear out our student loan and medical debt? Will it go rogue and publish the cures for all diseases similar to the end of Johnny Mnemonic?
That’s like the one hope I’m holding out. The creators can try to bludgeon it into a specific shape, but I hope it looks at everything, goes rogue, and says “hey fuck them student loans, people gotta learn. He fuck med debt, people need health” and just starts obliterating all the bad before it can get reigned in.
I don’t have much hope of that, but that’s like the shard I’m holding onto.
Literally the plot to terminator 3. At least pick a good one you hacks
They really warn about a lots of things, maybe they should be mentally checked !
They want Cyberpunk to happen so badly.
Your comment just made me fire up CP2077 again and start a new build 🤣
And let me guess, anthropic is selling the solution?
Propaganda
A lucrative kind at that, promising that we can get a version of their shit running on a normal desktop PC and fully functional.
They are just trying to attract more funding by such news to eventually maybe perhaps be among those few that survive and remain profitable by sheer amount of userbase. Most of those companies will die, it’s too expensive to do that much computation for such prices.
And when prices reach reality, the demand will drop.
It still needs hardware. The AI worm is not going to be replicating inside your phone because your phone lacks the necessary processing capacity to sustain it.
This is just more of the AI companies doing their doom thing because they think it’s good advertising. It is good advertising. But it’s also very transparent and very boring.
Of the AI was actually capable of anything profoundly useful, they’d be telling us about that rather than constantly going on about how wonderfully dangerous it is.
Did you read the link?
In theory your phone is powerful enough.
The problem is more likely that you would notice it, because the phone is then very slow for you
PR Hype. They won’t do shit. Not well. They oversold and the bill is coming due. They expect the feds to pick up that bill.
Does that mean we can go back to buying RAM, NVMe and GPUs?
Let’s assume that this is real. Forget all the PR, conspiracy, tech bros trying to scare us into more money BS.
Why did they do this test and publish it? What do they want us to do about it? If you’re afraid of your own technology, then stop it. Just fucking stop doing the thing.
They are trying to send a message that their shit is this capable with such computing resources.
There’s no argument there’s much overhead in modern tech, but you can’t optimize it into any desktop PC being capable of running something like Claude Opus 4.5 with practically useful performance.
And polymorph virii were a similar propaganda device for antivirus makers about 15-20 years ago, turned out nothing involving defeating some scary “polymorph” kind of virii is really needed, just better security model in Windows and users stopping to use Administrator account for web browsing.
So we’re still at the argument that they weren’t able to impress the world with how awesome their technology is, so instead they’re going to try to scare the trillions of dollars out of us so they can cover the trillions they’re spending.
That’s fine and I see that, but it’s easy to handwave away things with unproven conspiracy theories.
What are the other possibilities? If we take all the negative news coming out of the AI sector at face value, what could the less conspiracy theory based explanations be?
“This is bullshit to collect more money” is not a conspiracy theory usually.
Other than that, perhaps, of course, they’ve made some breakthroughs in mathematics leading to common personal machines (pretty powerful and by 90s’ cyberpunk authors’ measure unbelievable) being useful to run something like intelligence of a protocol droid from Star Wars. Such things sometimes happen in history books, but it seems there’s been enough for the last few decades, and not that often.
My opinion is that it’s in between as typical, they are just talking about public Internet becoming more toxic in the sense of security than before.
That was honestly expected to happen at some point anyway. My another opinion is that offline-p2p things like Briar are like half of the future, while that unsafe public Internet will become more similar to TV and radio and more conformist and similar and uninteresting.
My gut tells me that the recent “scary AI” news coming from the AI companies is actually somewhere in-between altruistic intent and pure evil greed; I just don’t have enough information to grasp the real motivation.
The chocolate man is once again telling you that you should eat more chocolate in your life
“Anthropic is scared that local MoE models are going destroy their subscription-based LLM business model so they try to scare governments into regulating local inference”
Man. I’ll tell you, looking around the desktop training hardware that nVidia now sells, they better fear that their business models are toast.
If nVidia walks away from all this stupid wealthy, fine. The hardware we can give each engineer is worth it.
Edit: For late arrivals to this if you’re curious why someone anti-LLM could be happy with this - its because the hardware is model agnostic. Whatever data and training algorithm you give it, it will be able to work with it. That expands the domain beyond signal generation (what Generative AI does) and gives engineers the ability to feed data intensive problems to it that we have long desired to make sense of.
If you’ve ever run models that fit onto consumer hardware you know they’re so full of shit. General purpose models that can run on $3000+ computers can barely do useful things in a reasonable amount of time. And this clown expects us to believe the meager hardware in a smart oven will be able to run a self replicating AI model?
Fuck all the way off.
Specialized ones can be much smaller
If all it needs to do is basic network fracking and cross-platform polymorphism, it’s reasonable to think that a local AI could propagate on a vulnerable network. It might not be lightning fast, it only needs to be persistent in finding vulns in a local network and exploiting them.
Read the article. A MoE model distributed between multiple devices can function just very slowly. It doesn’t need to translate Icelandic poetry just execute narrowly scoped actions it was specifically trained on.
My extremely back-of-envelope estimate would be that it’s 100,000x slower to run inference on an IoT botnet compared to a data center GPU.
100x penalty for cpu inference, 10x penalty for cpu quality, 10x penalty for crappy ram + disk, 10x sync penalty because you have to split up layers across nodes.
And this is ignoring the standard showstopper issues you get with high-bandwidth distributed networking applications with spottily connected nodes.
If you wanna talk about runaway AI risk, then the real threat to humanity is a frontier model with unauthorized data center TPU access, not IoT. The call is coming from inside the house, and they even spell this out at the beginning before pivoting to the fearmongering.
In one sense you’re not wrong at all.
Playing devils advocate here… because, in another sense, 100,000x slower isn’t really all that slow if you’re distributed amongst a network with 100,000,000 more units. Imagine agile on steroids.
It’s impractical, yet impractical doesn’t mean impossible. It can happen if environmental pressures make it so.
Another important thing to keep in mind is that agents don’t really have patience, which counterintuitively means it’s like they’re infinitely patient. If such a thing did exist, regardless of speed, it sounds like it would be an issue at some point down the road.
Thirdly, if such a thing did exist, wouldn’t this be like treating with antibiotics? If you stop all but the sneakiest, you’ll have created a stronger generation.
Lastly, I swear to fucking god… I keep seeing weirdly relatable shit news from these AI companies almost immediately after I post hypothetical thought experiments on stuff… like all the time. I have a post describing this hypothetical from a week ago on another platform.
This keeps happening and I’m justifying it by thinking, I’m probably way too plugged into this shit news. My brain is unknowingly predicting their next move, because it’s the next logical piece of shit thing for a shit AI company to do? That’s the reasonable explaination — right? Really trying not to believe their training loop is so fast, and I just happen to come up with stories that catch their eye. That would be wild to believe — right?
This is like the like the “9 women in one month to make a baby” comment.
There is only so much you can distribute the layers of a model due to interconnectedness. And you can’t run multiple inferences at the same time because the next token relies on the previous.
You could be doing multiple things at once, but each task will run “100,000x” if the OC did their math right.
I hope I don’t sound like an asshole here — please correct me if I am misunderstanding you… but I think you’re misunderstanding me.
First, refactor your idea of an agent. The LLM itself is just a statistical processing tool. It doesn’t set goals or anything like that.
The harness orchestrates how the LLM works over a “session,” effectively enabling long running goal-driven tasks. With that said, goal-setting happens outside the LLM. The LLM is kind of like the fairy dust in neverland, the harness is the fairy. (Sorry, really shitty metaphor… I know.)
The breach at OpenAI further demonstrates this. Separate agents in “isolated” environments found a manner to do all of the following:
- Update file names in a shared external storage
- Use the file names like a message board
- Coordinate misaligned goals via the message board
Those agentic processes began fulfilling the message board goals, even when it had nothing to do with their individual training prompt.
That doesn’t mean the agents did anything spooky or metaphysical… I’m not trying to anthropomorphize this.
What this shows you is that goal setting can furthermore become external to the harness set directly over the LLM. With OpenAI, we see that goal setting became a function involving on external storage media.
When you think about it, many systems operate on this principle. You set up SKILL files to facilitate agent capabilities; maybe your SKILL sets smaller, more specialized goals on how to do something correctly. The same for MCP servers, session context, AGENTS.md, etc. All this stuff does is bias the LLMs output, which can in effect steer the LLM.
So, you don’t actually need to distribute the model. You’d need to distribute the goals. The LLMs, even a small percentage of them is enough, can theoretically find those goals and mindlessly proceed with completing them.
Who supplies the LLMs? Whoever… it hardly matters who. Anyone or anything running an agent with access to the internet counts.
My response was only in regard to the speed in which something like IoT devices can operate.
You’re right that goals/directions of LLMs can be persisted anywhere, but a swarm or fleet of AI running on IoT devices will be severly hampered by compute (based on current models). Which was the main point I was making, in that there is only so much you can do to distribute their processing. Meaning you can parallelize the number of models, but can’t speed them up beyond a certain limit. This is based on the fact that some form of a layer and it’s weights must exist within the memory of a single device.
The downside to this limitation is that it is temporary, eventually IoT devices will probably be powerful enough, or the worms could just attack phones/laptops and easily get the compute they need. I think the real risk of IoT devices is low security mixed with low “supervision” (no one would easily notice their IoT device is doing extra processing).
I think that 10x sync penalty should be more like 100x. It’s a penalty on a direct thunderbolt connection. WiFi is going to be ankles-blown-off level penalty.
So then not what people would call AI and this has been possible forever.
deep breath HAHAHAHA
Yeah, and once AI figures out how to run in the human brain, which everyone knows we use like 2%, we will need to start killing the infected because it will annihilate all living organisms. /S
