I don't think he ever said in the article that he is not making the argument about the importance of democratic oversight, if anything, in the conclusion, he says:
"The way to address this new reality, however, is with new laws and through strengthening accountable oversight; cheering or even demanding that an unelected executive decide how and where such powerful capabilities can be used is the road an even more despotic future."
But I think you drastically misunderstood the point of this article. Ben is pointing out the implications of Amodei's analogy of advanced AI being like nuclear weapons. The government has a monopoly on nuclear weapons and has extreme regulations and oversight on the companies that help build nuclear weapons for it. And those companies do not tell the government how or when they can use the nukes.
So if advanced AI is like nuclear weapons, why can an unelected executive tell a democratically elected government how to use it?
This analogy makes no sense. You and I can go use Claude right now. The government clearly doesn't think they are nukes. If the government wants to control this technology like a weapon then it should do that, but that's not what's happening.
That's not what is happening, yet. If this technology is at the scale of nuclear weapons, you don't think the government is going to step in and take control of it?
Potentially. Or the US government can force over control of the technology. It's impossible to predict right now the implications of this technology in 5 years, let alone 10 years.
You’re right, the analogy makes no sense. And what’s more, Dario didn’t make the analogy. This blogger took a partial quote out of context and constructed a blog post around it.
If you listen to the interview, Dario is talking about the Trump trade deal with Nvidia being allowed to sell H200 chips to China, but the USA gets some taxes. He thinks it’s a bad deal for long term national security. He likens it to “selling NK nuclear weapons and bragging that the casings are made by Boeing”.
It’s not about AI/nukes being equivalents- it’s about making a bad deal with a rival, giving them something they want / reducing your superiority, just so you can brag about having made a deal.
I'm not well-read on the history of nuclear, but is it the case that nuclear developed first in government and was then spread to private industry via gov't outreach/motivation?
Here, I think we're talking about the opposite, right? Private developed, then gov't used. It's so obvious in the first path that gov't would remain in the control, but I'm not sure how to think about what's "right" in the private-to-gov't path.
You're absolutely right and the order of operations matters. But I'm not arguing about what's right in this case. If as Dario says, that advanced AI is at the level of nuclear weapons, then governments around the world will see it as a threat to their power and sovereignty.
Govt creates something: you see that tight control from the beginning. Makes logical sense.
Private company creates something gov't sees as a threat: private companies are allowed to express opinions and make decisions irrespective of threat or not, right?
One example I'm thinking of: Apple's stand on cracking devices. Gov't wants to crack into a phone and Apple won't do it for them without right measures in place. I think Apple SHOULD be able to make that call. Whether it's right or wrong is still murky.
If Ford sells Broncos, and Pete Hegseth says “I’m Batman, we want them with RPG launchers”, Ford is not required to create Bronco’s with RPG launchers.
If Pete wants AI killer robots and AI domestic mass surveillance tools, he can go put out a RFQ like literally any other DOD DARPA project in history and get bids.
You're right in those cases. But if what Dario says is true, that advanced AI is on the scale of nuclear weapons, then that is a threat to the power and sovereignty of the government.
That’s not what Dario said. This guy Ben took a partial quote and then used it to construct a strawman. And hey look, he was successful, you bought it.
Dario criticized the Trump admin policy of allowing Nvidia to sell advanced chipsets to China as being poor for long term national security. He said our model edge is dependent on the HW edge, and likened this Trump “deal” to selling nuclear weapons to NK and bragging that the casings were made by Boeing. You can go watch the video yourself.
He was attacking the Trump trade deal as being bad, and characterizing Trump as more concerned with being able to name drop a US company and say he made a great deal, rather than actually doing what is in the best interest of the country.
Do you just believe everything you read on the internet? Or was it Ben the bloggers amazing credentials of being an Apple INTERN in the company culture group? He’s a real philosopher king I’m sure.
But that's the problem with this. The vast majority of the spend is going to be on the Nvidia chips which have a shelf life of 3-5 years. They are not making any significant long term investments.
During the dot com bubble, telecom companies spent 10s of billions of dollars laying down cables and building out the modern public internet infrastructure that we are still using today. Even if a lot of companies failed, we still greatly benefited from some of the the investments they made.
For this bubble, the only long term investment benefits seems to be the electricity build out and a renewed interest and investment in nuclear.
Most (if not all) of Oracle's investments are mostly in chips and data centers.
I was going to comment this exact thing about stacking plates. I think most servers/ex-servers also do this regardless of age. It's even easier to do than returning a shopping cart.
The only downside of this in the US is that homeless people will tend to hang around Aldi's asking people if they can return their cart to get the coin. Most of them are friendly and thankful but every once in awhile an aggressive person would make me very uncomfortable.
I also expect Aldi management isn't thrilled about homeless people camping outside their stores.
It already has. Any tech company that is pre-IPO and still raising funding rounds is a "startup". I'm surprised there hasn't been someone to come up with a separate term for the stage of these kinds of companies.
Still, I think there needs to be a specific term for a company that has recently had a funding round and will likely IPO in the future, like Stripe. That's a different category than a start-up or privately owned company that will never IPO like Koch Inc.
I think it makes sense in this instance. Because this occurred in us-east-1, the vast majority of affected customers are US based. For most people, it's easier to do the timezone conversion from PT than UTC.
us-east-1 is an exceptional Amazon region; it hosts many global services as well as services which are not yet available in other regions. Most AWS customers worldwide probably have an indirect dependency on us-east-1.
I think what you're trying to reference is APIs or libraries, most of which I wouldn't consider abstractions. I would hope most senior front-end developers are capable of developing a date library for their use case, but in almost all cases it's better to use the built in Date class, moment, etc. But that's not an abstraction.
Because if LLM inference is going to be a bigger priority for the majority of companies, they're going to go where they can get the best performance to cost ratio. AWS is falling behind on this. So companies (especially new ones) are going to start using GCP or Azure, and if they're already there for their LLM workloads, why not run the rest of the infrastructure there?
It's similar to how AWS became the de-facto cloud provider for newer companies. They struggled to convince existing Microsoft shops to migrate to AWS, instead most of the companies just migrated to Azure. If LLMs/AI become a major factor in new companies deciding which will be their default cloud provider, they're going to pick GCP or Azure.
reply