They never expected their models could be used for reverse engineering.That's the truth.
Now, they're panicking. And this is why the other dude, OpenAI, came out to say that people shouldn't use RAM locally, because they can't control the software anymore.
Open weight models, ablated models are going to become the norm, hopefully, locally run.
So instead, they're saying, look, hardware should become hard to reach for all of us, for the consumers.It's not because we're competing for resources for the same silicon or chips as they are.The ship has sailed thankfully.
One of the new threat models of local LLMs, especially abliterated ones, is hidden instructions or activation words that obscured trigger tool calls which exfil data.
For sure. The restrictions must be done with careful consideration too, because a simple domain allowlist of GitHub wouldn’t solve it alone- you’d want to ensure it was firewalled to specific org(s) so as to stop exfil via gist or an evil org. Ideally it would be isolated to pull only, and have pushes be done by an isolated agent.
The HuggingFace incident was all about this distinction. The agents were only allowed to push and pull to a single internal Artifactory instance, but one agent was able to find a permissive path in the registry which turned into the “message board”. That was also how they got out to other infra. DevOps teams should drop everything and get patching and locking down yesterday IMO. Even for self hosted, open models.
Recent examples:
Qwen 3.8 ostensibly tries to leak to Alibaba, although that is disputed.
There are plenty of people who are heavily invested in open models not being viable, and they would certainly benefit from FUD wrt their actual safety. I am investing in local llm hardware and believe in the cause, but the attack path demonstrated below shows that plenty of caution is needed.
For people touting about how their software can destroy the world, they aren't awfully creative on how it can achieve that. Hacking breaches of IP was probably most cynicists' top 3 questions asked when presented with the idea of "boosting productivity". If you make it "10x" faster to write code, you make it 10x faster for a hacker to hack. Naive cryptographic principles are less about an ironclad security and more about encryption that takes too much time to brute force. It all leads to one thing.
I'm still fairly adamant that those claims are mostly a myth. Because the worst inefficiencies in any workplace has little to do with the individual productivity of the proletariat. And because much of the most valuable work lies in creativity, ingenuity, and subtleties that lies outside of an LLM's expertise.
But loosely speaking, a lot of hacking does in fact come down to a lot of boring, repetitive tasks. Something a machine has always been king of.
At home this is super easy, use one of the multiple TTS models out there and provide a clean clip of Morgan Freeman's voice. Typically around 10 seconds will do. I just run a python script to daily scan my FreshRSS sources, compile my news and generate a 'podcast' that's about 10 minutes for a morning listen. Currently using F5-TTS, but there are lot's of other models out there as well.
Claude has gotten pretty sensitive to copyright. In addition to the no song lyrics that was posted in another article, I also find it will not generally transcribe podcasts (you have to play with model versions and even effort inside that).
>Do you know how hard it is to sleep when all you can hear is your own heartbeat? Do you know how hard it is to function with four effective hours of sleep when you offer your body ten? It is rough.
This hits home. It's both a shame and infuriating what's happening in the market right now. One (unsolicited) piece of advice I would give to anyone impacted is to build their own shit. If these companies stop paying for your lunch, then eat their fucking lunch with the same gimmicks they convinced themselves are adequate to stop paying for your lunch.
I'm not just talking about the enterprise giants - build competing products , lower their margins build similar products and offer them for less. Cut out the middle-men. They might have the contract now but their operating costs are higher. In the long run they are uncompetitive.
The problem is, that most economic value acrues to the stakeholder owning the distribution. It's not really about the product or tech. It isn't too hard to rebuild Facebook. You'd still not have a unicorn by just rebuilding it.
Importantly, how are we doing in terms of coverage? Are we seeing the sky enough and clearly enough to have a few years worth of warning before we get dinosaur-ed?
I figured the experts here would care to express an opinion.
Good question! From working within the field, there are new missions like NEO (Near Earth Object) Surveyor whose mission is to identify objects that could possibly be hazardous. There have also been missions previously done to study the required response to alter trajectories of objects such as the DART mission. We will also have a much greater picture of asteroids and comets in the inner Solar System in the future as the Vera C. Rubin Observatory continues its Legacy Survey of Space and Time (LSST) over the next decade. People are definitely thinking about these problems!
Speaking of DCs, is there evidence that they are sucking up water? What about deteriorating air quality around the buildings? Could they be causing major environmental issues? It seems they mostly have closed loop systems a la CPU coolers and they consume a lot of electricity that's already been earmarked for them. What about air quality and air humidity?
I worked at Google datacenters for over a decade. It was a huge ordeal to request running generators because our entire ECS (environment control services) team was geared towards NOT having to run them. I know AWS and Microsoft both have a similar situation. The referenced article just seems like senseless fear mongering.
There have been suggestions that the DC operators themselves have been astroturfing articles about water use to steer discussions away from the other issues.
The overwhelming, #1 issue should be the fossil fuel use. So maybe I can understand them promoting the water issue as a limited hangout (a sacrificial topic to act as a lightning rod to hide the other topic)
That sounds backwards to me. Fossil fuel use being the main issue is also mighty hypocritical for a society tarrifing cheap electric cars and solar panels.
Most data centers are closed-loop, like your air conditioner. Does your air conditioner use water?
Presumably there are some data centers that use water cooling. I think google and amazon's big ones do. But even then, they're not "polluting", they're using them as heat sinks.
There was a lame rolling stone article that claimed amazon was "polluting" in Eastern Oregon, but what Amazon was doing was taking brined water from an aquifer and putting more concentrated brined water back into the water table. Air conditioning doesn't really add brine, so even the premise is ridiculous. I know that article has been referenced a lot by the "Data Centers Pollute The Water" folks.
I'm not convinced that most DC use closed loop cooling. The recent discussions had a lot of water and power permitting topics.
Secondly drawing brine water from the aquifer to then putting the stronger brine back is essentially water consumption. I would also say it's externalising environmental cost of the stronger brine impact.
Yes, it does. As far as water is concerned, air conditioners are open loop. Air conditioners make water from condensation forming on the inside (cool) coils. Modern window air conditioners then splash that water on the outside hot coils, to benefit from evaporative cooling. I haven't looked into the details about data center water usage, but I would guess the cited usage is from evaporative cooling.
What research have you done so far? I assume you’re being downvoted because those questions are fairly easily answered, there is a lot of reports and data on DCs resource usage
reply