Do you mean to ask why non-American companies would trust anything coming out of the USA? Because American companies tend to have a lot of trust for other American companies.
Wow, TIL. Is there any calculator or rule of thumb you use to figure out the expected performance of a model on some DRAM/SSD split? For instance I am wondering what the performance could be like on a 512GB M5 Mac Studio.
The rule of thumb is to just test it on your chosen platform, with your workload. SSD performance matters a lot obviously, you'll want to run PCIe 5.0 if at all feasible, or a recent M5 Pro/Max/Ultra or M6 Mac with high storage capacity (lower tiers have much worse performance). You can figure out a worst case easily (assuming that the fraction of sparse params you manage to cache in VRAM or system RAM has a fair chance of being picked among the total) but in practice you do quite a bit better than that - which is why even very small caches can be potentially worthwhile.
Ok, but Hinton is worried about what happens if the tech keeps getting better at the current pace.
Let's flip your last point: if the tech gets good enough to replace people at any task (including physical ones), do you think others will need you for much?
And then there's the systems that the corporations are making more powerful as fast as they can. Yes, they promise to do things like cure cancer, but they're also capable of scarier things like hacking banks. And they're progressing faster than we can understand them. Would it be so bad to delay the cancer cures by a couple years to ensure we still understand what the models are doing?
With AI-driven speech recognition, I have far more agency than I did without, period. I can write, I can code, I can communicate with various entities, people, government organizations, corporations, chatbots. I can fill in online forms instead of leaving a bunch of spastic scrawls that generate more questions. I can use it to drive OpenSCAD and 3d print something.
And here's why I start out by saying I don't care if others need me.
I think I'm congenitally incapable of not building/designing/exploring things. If we had an AI-dominated future and we were all on UBI, I'd still find myself saying “hmmm, I wonder” and going down a rat hole to see what I can find. I'm doing it for myself. My curiosity and satisfaction in bringing something to life, even if it's only a variation on what somebody else did.
Rephrasing something I said in a comment somewhere that I would spend time gardening, bird watching, and building a closer relationship with my partner and my pets. I would have no want of things to do or explore.
To your last point, we don't need to understand how something works for it to be useful. Doctors don't understand why treatments work and why they fail. If doctors waited, if treatments waited until they understood how and why it worked, I'd probably be dead. But because they don't, I now have a VP shunt in my head (aka, medical-grade juice-box straw), and my life has drastically improved. Toddlers are something parents don't understand—the how or why they do what they do—but they keep having them anyway.
WTF moments are an emergent property of toddlers, goats, border collies, working dogs, hunting dogs, and AI agents. And like with a toddler, we need to deal with it like we're the adults in the room.
Too many people in my life have died of cancer, and the agony and pain as their body hollows out to a waxen shell is something I'll probably never forget. If I had a choice between understanding how AI works and a cancer cure, I would choose the cancer cure, hands down.
One thing we do need to do is slow down the development or the build-out of AI data centers and put a resource constraint on the model companies, forcing them to get better with the same resources. Sort of an economic “you can't have dessert until you've eaten your vegetables” moment.
I really got to sit down and write that blog post on why agents are like toddlers and that toddlers should have figurative machine guns.
Without accessibility tools, there's not much incremental value I can provide. When the RSI showed up, I was advised by many people to spend the rest of my life on SSDI. I refuse to do that, but it was difficult to show I could provide incremental value. Yes, my mind still worked; the designer, the architect were still there. I had speech recognition so I could write, but a common reaction was, “How can you do anything technical? Your hands don't work.”
Now to the general case that you seem to be worried about: that AI will remove everybody's ability to provide incremental value. No worries there. The entire system will collapse. Labor has no value. Without labor, capital has no value. And we're all jolly well fucked. Now, Your value is how good you can become at asymmetrical class warfare.
At least that's where the current nihilist fantasies about AI leave you.
The AI nihilist view works when you have the wrong enemy. AI is merely the puppet, the distraction, just like the H-1B visa, offshoring, undocumented workers, etc. All these distractions are so that we won't notice the billionaires are enshittifying society and won't seek to dethrone them.
IMO, I argue the fix involves returning the tools of production to the hands of labor, and taking power away from capital. But as with many problems in life, the solution is left as an exercise for the reader.
I'm confused, why does capital have no value without labor? I thought the whole point of AI is to enable capital to fully replace labor.
I sympathize with "dethroning the billionaires" but to me it looks a lot like powerful, general AI systems are the most capital enhancing technology we've ever seen. Wouldn't the class struggle benefit from slowing down frontier AI development while diffusing local models to ordinary people? I feel both can happen at the same time.
Those are good questions. IMO, capital only has value when it can be exchanged for an asset. Assets are the direct result of labor or hallucinations, e.g., a brick or a share of stock. If there's no labor, there are no more assets produced. If there are no assets, the value of capital goes down. Occasionally I think the endgame is reducing everyone to the oldest and second oldest professions.
As to the point of AI, it's fuzzy. We have a tool we don't understand how to use; it's incredibly valuable in some spaces, destructive in others, but we really don't know how or where to use it. We are trying to fit it onto the old model of work and labor value creation but don't yet understand what the new model should be. It's analogous to generals fighting the last war.
General AI systems: capital enhancing? I can understand how you and many others see that, but right now it is capital-consuming. Capitol is desperately looking for that model they can wield to get more capital out of the system than they put in. With the opinion reports describing frontier AI companies' bond offerings as junk, it looks like asset gamblers with capital are trying to unwind their exposure because they don't see how they're going to get returns.
I suspect the meme of using AI to reduce or eliminate labor is a continuation of capital's hostility to labor having any power at all. It has traction because capital sees reducing labor costs as a way of extracting value from their investments, and, as I said, they can't see any other way to do it.
Slowing down development won't make a difference to the class struggle. There are many parties in this race, and handicapping one does not slow down the others. However, slowing down deployment, taking time to figure out the right way to use it would greatly benefit the class struggle. Denying AI companies' data centers is probably the most effective way to slow them down. If they don't have the resources to spend tokens on answering your email, they will find more valuable ways to spend tokens that should benefit all of us.
Denying them data centers will free up resources, which will enable us to create distributed AI use. Altman's view that all the RAM belongs to him because it's more efficient is a valid model if you are trying to fit AI into the way we used to do things. Progress and new economic development seem to emerge from the edges and not the central power structures of an economy. So if we want to find that new model of how to work with AI, it is imperative that we can build out local models in the hands of many people.
No, the AI compute supply chain is actually quite similar to the nuclear supply chain. Building chips capable of training or large-scale inference is extraordinarily complex, expensive, has a large footprint, and has a few chokepoints (rare earths mining, HBM manufacturing, and ultraviolet lithography, where there is literally one company in the world capable of building the relevant outputs) where we in fact already have governance mechanisms for things like export controls.
I wouldn't go that far. These companies are leaky and there can be a backlash that causes people to lose trust and move to competitors. For instance, when people discovered the Grok CLI was uploading repositories to the cloud despite the training toggle being disabled.
Ok, but all of these technologies have bounded impact, and they present new abstractions that need to be learned by engineers.
Coding agents are different in that they are unbounded in the tasks they can do (although reliability is an issue), and you can work with them the same way you work with other engineers (e.g. via natural language).
Also, coding agent capabilities are advancing much faster than people can sit down and learn their limitations. All the tricks we learned for Opus 5 became obsolete in a couple months with Opus 5.5.
> Stanford researchers find the employment trends they identified for entry-level workers last year are persisting and expanding. Specifically, employment levels for workers ages 22 to 25 in the most “AI-exposed” occupations are now 19 percent below those of their peers in fields less exposed to AI disruption.
I feel this is a common position from people who do not want to acknowledge that society and people's ways of life will profoundly change. It's a scary proposition so I don't blame anyone for it.
I used to think that it is very important to argue and convince people who are in denial so that they can e.g. prepare for it. But nowadays I have been thinking that being convinced of this scary proposition may also lead people to behave in destructive ways (also, I would like for us to spend less on AI, perhaps because it is so powerful), so maybe it's not so bad.
reply