This will never be worth the risk. Every plumbing job is a unique challenge, you can't just spin up 10000 agents to brute force your way to a correctly cut pipe, one mistake and you've destroyed a house or office.
No my point is that free speech isn't a spectrum, we either can say whatever we want or we can't. As far as whether we have it, no in the US we don't have it today as certain speech is still made illegal.
I don't cast free speech in absolute terms, I take issue with those that do since it isn't a spectrum.
Free speech isn't a spectrum in the same way altruism isn't a spectrum. We probably all agree there is no true speech, so we change "free" (as in true freedom) to "free" (as in "liberty") and establish a spectrum anyway.
No one wants "true" free speech anyway. an age of social media + deepfakes means true free speech is legalizing slander/libel. And we know the lies will persist much longer than it deserves.
As well as inciting violence. We don't even need to reach "hate speech" on the spectrum in order to talk about intolerable speech everyone agrees should be punished on.
I'm impressed that you can speak for everyone collectively here.
I take issue with any law that limits speech, including libel, regardless of who says it. Such laws remove agency, and therefore a level of responsibility, from those who hear the speech and choose to act on it.
In the same way that I strongly disagree with laying any blame on a woman for dressing a certain way before being attacked, I don't blame a person for saying something before another person takes an action we've deemed illegal.
> RSI is the play and it’s already happening. Our work is easier and faster and anyone that claims otherwise I have a very hard time taking serious.
No it isn't. Doing work more efficiently through the use of LLMs is not RSI. RSI requires a closed loop automated LLM iteration process which we are not at all close to right now.
Do the air bags in your car go off at random while you're in traffic? If not, then why do you think a system using these jets to avoid accidents would?
Also where does the habit of ending angry messages with lol or lmao come from, which I only see here on HN?
"After the air bags were replaced and the vehicle was involved in a crash, the air bag inflators malfunctioned and ruptured, sending large metal fragments into drivers’ chests, necks, eyes and faces, leading to death or severe injury."
So yes, safety features sometimes have flaws.
> If not, then why do you think a system using these jets to avoid accidents would?
For starters, airbags are reactive to an impact, not predictive in anticipation of it.
"We're going to crash" is harder to assess than "hey a crumple sensor in the front bumper just got wiped out".
(The airbag is also deeply unlikely to obliterate a different car, pedestrian, or bike during deployment. They're simply not comparable to a rocket engine on the outside of the car!)
And no one was using AI for serious work a few years ago. Technology develops and moves forward. Practically every new car and motorcycle today has safety features which depend on computers. And they function in practice.
Trucks all have pneumatic brakes and an air compressor on-board. So that part is nothing new. A system of air jets could possibly be implemented as a correcting measure if a truck is overturning. There's no AI needed for the truck to know it's about to overturn. And trucks overturning are also some of the most serious and most dangerous common accidents in traffic.
But self proclaimed "hackers" are too busy with LMFAOing and LOLing at any suggestion or speculation on how technology could be useful in practice.
> Also where does the habit of ending angry messages with lol
I'm using lol here as an expression of rueful delight, not anger. I'm not being sarcastic - the mental image of vehicles autonomously leaping through traffic is really funny to me. However, I'm slightly perturbed by the lack of parsimonious thinking.
> When I retrain its original architecture, which still produces messy deep layers, and follow the least legible neurons I can find, they turn out to be overgrowing locally into reasonable, adjacent territory.
It's interesting to observe that human reasoning is still the barometer of what "reasonably adjacent" means in the context of a NN's structure.
The fact that they're using Artifactory in the first place is already a failure. The fact that they didn't airgap the network or even use any type of egress blocking is just absurd.
Zero days are nothing new, secure containment has been a solved problem for a long time.
I'm curious how you would airgap a datacenter-scale training run. No reading training data from a cloud store, someone from OpenAI flies out to Abilene or whereever with a suitcase full of SSDs? When researchers in SF write code for a new RL environment, they ping somebody who sticks it on a flash drive and walks over to the server rack with the GPUs? Has anyone ever trained an LLM like this?
Maybe you don't mean a literal airgap, you just mean configuring Internet access off. But... they did that. That's why the agents had to hack Artifactory to get it...
Neat, never heard of those. I'm curious how this works with, like, acks, and also whether there's any public prior art for anyone having trained an LLM beyond one.
As someone who has actually had to maintain multiple airgapped networks in the past and knows the amount of work that's required for that, especially at the scales of data they're working with...
All the armchair cybersec experts claiming they "obviously" should have airgapped are the ones being absurd. Especially in the same breath as claiming the models apparently aren't intelligent enough to warrant such concern.
Nah. It really wouldn't be hard to set up an environment where models can't damage the outside world from. The fact that people still don't think this should be the norm or paint it as some technical impossibility is mindblowing.
People execute, study, develop, and test literal malware of every caliber everyday without issues. Yes the way they do it is requires care and in restricted settings. Yes it costs money and takes time to do that. Yes the people who do it are required and expected to do that.
Oai and anthropic are more or less run by clowns who intend to profiteer off of regulatory capture and who knows maybe directly from the harm their technologies can cause.Because of these takes we have to wait while they try to expand their verticals into the pain and suffering industries. Palantir is already pitching smoking cigarettes so there's that.
Well, as someone who has had to maintain multiple such airgapped networks you are obviously in the know that this could be done, because you've done it before.
The point here is: they didn't even try.
They could have easily temporarily cut off the segment that this was running on, or restricted and monitored its egress to the point that a hacking attempt on an outside party would have been detected.
The point here is (again): they didn't even try.
So yes, they obviously should have either done that or they should have not run their experiment given that they are claiming that the models are intelligent enough warrant such concern. In fact they post one article after another about how much smaller and less capable models are intelligent enough to warrant such concern and in fact warrant regulating this domain to the point that only they are able to own and operate this stuff.
Let's make a biology analogy: if you can't safely operate with harmful substances that doesn't give you a license to do so in an unsafe way. Then you just don't do it.
You misunderstand me. The work isn't primarily in "securing" the network. It's the hassle involved in the continuous transfer of data from one side to the other (and for a proper air-gap, that transfer only goes one way, which is laughably infeasible here). In my line of work that's mostly software security patches and weekly code transfers, but obviously these labs have multiple orders of magnitude more.
And I don't know how you expect an LLM to do airgapped data transfers. The fact that you even bring that up makes me think you don't actually understand what airgapping entails.
I'm no expert, but wouldn't a basic hardware firewall that allows incoming, but not outgoing connections solve the issue? Wouldn't even basic VLAN isolation achieve the same?
The humans can still open a TCP connection, SSH tunnel, etc. from the outside to remotely get data out of the training environment. But the agents cannot reach anything outside their isolated network.
Yes it would. It's not perfect like a literal air-gap, but it's the bare minimum. Had the agents breached a real, properly configured firewall product, that would've been a real watershed moment.
Things that are not intelligent could still warrant very strong safety measures.
Depending on the risks associated, the complexity and effort of certain containment measures would also not neccessarily be something that could be used to brush them aside.
These companies are taking millions of dollars, sucking in experts everywhere they can, and that is before they even hit the stock market.
I don't think it is too much to ask that they devote their resources to creating an air gaped campus for testing. If they need to build an entire data center campus with copies of the pilfered data they used to train the damn things, then so be it. HF and other incidents have proven that trying to block models off by software networking is a dangerous game when they can just exploit some unknown flaw in a library nobody knew about for over 40 years.
The inconvenience of implementing correct safety measures is part of the cost of undertaking certain activities. When an activity's ratio of cost versus benefit is too high, there's no way to juice the benefit, and the last place to cost-cut is in safety measures, the responsible thing is to refrain from the activity in the first place.
reply