HN user
I would like to see a citation on that b/c I am unaware of it. The only case I see in SDNY is the NYT v OpenAI case which has not been ruled on yet. https://www.reuters.com/legal/legalindustry/copyright-law-20...
You are asking to prove a negative. But even assuming that the model is capable of returning every bit of its training data verbatim (a mathematical impossibility) that would not be enough as mere capability is insufficient here. If capability alone were the standard any library that also has a photocopier / scanner would be in violation.
To prove distribution of copyrighted materials it would have to be practical and actually used in the wild by people to circumvent copyright and generate copies of those works. Again, I can't prove a negative, but that isn't the standard, and nobody has shown a practical exploit here.
That is a false statement. Gaining commercial advantage means selling pirated copies which Anthropic absolutely did not do, so none of your following statements are correct either.
1/4 resolution, but still 100% of the movie. There's not really an equivalent for a book.
A model absolutely could be trained to engage in malicious behavior like that, but it seems impractical for an actual attack. What you want as an attacker is to insert a backdoor exactly where you want it an not where you don't because every backdoor increases your chance of getting caught. A malicious model inserts backdoors and exfiltrates data everywhere and you care about maybe 0.01% of it. The other 99.99% is negative value to you. In practice this malicious model would be caught almost instantly.
A hosted model is different because you could prompt inject specific customers, but I assume from this question you mean a malicious open source model being hosted by an honest provider.
Biased sample, but basically everyone I know on either side of politics agrees with that.
Well you'll need to keep waiting because I never made such a claim and therefore I have no need to justify it.
Residential electricity rates don't get "bid up" like that. Increases typically must be approved by the government and have to be justified by expenses rather than market prices. Ashburn VA, which has more data centers than anywhere else, actually pays below US average rates.
All of the above are false, except for 3 which has actually happened in a few cases, though is definitely the outlier as I have been near data centers and couldn't hear a thing. And they provide a lot more jobs than the 0 jobs that were provided by the undeveloped land they are built on.
I think you have made up a strawman here. Nobody actually thinks people just decide one day "I want to be homeless." The conservative position is they are homeless due to their own bad choices, which is absolutely true in a lot of cases. Their error comes from assuming that they can just reverse the situation by not making those bad choices anymore, but in reality it can be a one way door.
If you want dumb conservative positions on homelessness I would suggest "just lock them all up", "just make them get jobs", "just get them out of MY neighborhood."
A v2 Starlink satellite costs $800K and on average 25 are launched at once. Launch cost for a reusable Falcon 9 is $15 million. So that's $1.4 million per satellite to orbit lasting 5 years that's $280K / sat / y, or $2.8 billion / y to maintain a constellation of 10,000. And SpaceX is not known for complacency. The unit cost will continue to drop.
On the other hand there are currently $63 billion (22.5 years of Starlink cost) of rural broadband subsidies active in the US and it hasn't come close to running all that fiber. So $63 billion to not even finish the US vs $2.8b / y to provide service to the entire world. I think it's safe to conclude that the satellite option is in fact much cheaper.
I suspect what is going on is just a matter of relative density. I'm not sure what you mean exactly by "central EU," but just guessing from a map I get Romania as the least population dense country that I would think of as Central Europe at 83 / km3. That is more than double the US pop density and if it were a US state only 15 out of 50 would be more dense. So then taking the least population dense region of the least dense country I get Tulcea with 23 / km3. That's 66% of the density of the US (37) which would come in at 34 / 50 if it were a US state.
So the most sparsely populated region of the most sparsely populated country in Central Europe is just a bit below average for the US. Our least dense state is Alaska at 0.5 / km3 or almost 50x less dense than that. But that's almost cheating. So lets take mainland only and that's Wyoming, with 2.3, so 10x less densely populated than the outlier in Central Europe.
So basically the US is just really damn empty to the point there just isn't any comparison in Central Europe and that's why it's so hard to get internet access out there.
Correct, but it goes deeper than just the building components. In the US you have to go through an entire military procurement process within each iteration loop. So you design a weapon, then try to sell it to the military, but just the process of demonstrating it and selling it to the military takes a long time and costs money. If you fail you can go back to the drawing board, but each iteration loop is probably a year minimum. And if you are successful now you have to set up and scale production. Get ready for years of environmental reviews and lawsuits.
In Ukraine the military will take any drone they can get their hands on, so all you have to do is build a drone, give a bunch of them to the army to try out on the Russians, and within a week they will tell you if it works or not. So your design iteration loop is probably weeks. If you are successful, the time between hearing the general say "give me 1 million" and when the bulldozers start clearing the factory site is probably measured in days.
And yet the regulation actually will result in more dead soldiers or else it wouldn't be the first thing that goes out the window in a war.
Should be a separate [stupid] flag, though.
There are plenty of things to complain about here, and that is one of them. But that authorization was passed by our elected representatives by a super majority and reauthorized by them multiple times. It was not done by a sneaky maneuver where the majority of congress voted against it but somehow it still became law.
Actual workers should outnumber "governance" bureaucrats by 100:1
I get it every time I touch the login path
And why am I supposed to care? Because plutonium sounds scary?
Which is precisely what government bureaucrats value your time in complying with their arbitrary bullshit at.
Why can't you? All other forms of power generation do that.
99.99% of the radiation is gone after 300 years, so you don't really have to.
By my personal standards for the term, the majority of philosophers don't have PhDs and the vast majority of philosophy PhDs are not philosophers.
One country is not in any way obligated or expected to have the same entry requirements as another. It is based on priorities of the state and those differ greatly between states.
Permanent residency is a deal (not necessarily business related) between two entities: an individual and a state. It has everything to do with whatever requirements that the two parties have, and if there is no agreement on them there is no deal. In this case the state cares about language proficiency and requires it for a deal, so if you are not proficient in German there is no deal.
Philosophy majors. That piece of paper does not make you a philosopher.
The strange part is that they seemed to have tricked AI companies too.
They sure haven't gone out of fashion on the streets of SF. My ears were ringing!
You're not the only one. There are few things I hate more than safetyists.
Get the drone show up and open it up to amateurs trying to shoot down the drone show for maximum adversarial drone warfare preparedness!