Yeah, I don't think the "release valve" is the correct metaphor. This is more like a crack around a door frame that you can get a lever into in order to eventually pry it open.
> that enables a computer to pretty effectively understand natural language
I'd argue that it pretty effectively mimics natural language. I don't think it really understands anything, it is just the best madlibs generator that the world has ever seen.
For many tasks, this is accurate 99+% of the time, and the failure cases may not matter. Most humans don't perform any better, and arguably regurgitate words without understanding as well.
But if the failure cases matter, then there is no actual understanding and the language the model is generating isn't ever getting "marked to market/reality" because there's no mental world model to check against. That isn't going to be usable if there are real-world consequences of the LLM getting things wrong, and they can wind up making very basic mistakes that humans wouldn't make--because we can innately understand how the world works and aren't always just stringing words together that sound good.
Break the aspects of language understanding and language generation apart. While I would agree that generative LLMs are understanding-free madlibs for writing text, embedding vector spaces and LLM latent spaces seem are a pretty genuine understanding of natural language. High dimensional vector spaces seem like the best machine representation we currently have for meaning and LLMs are using it effectively.
I've been hearing that China will hit a plateau, like Japan, for at least 20 years now... Meanwhile, China is now pumping out BEV trucks, affordable electric cars, sixth generation military jets, and nuclear aircraft carriers.
> China is now pumping out ... sixth generation military jets, and nuclear aircraft carriers.
They are not. They have some initial versions.
The thing about long distance navies is that they require lots and lots of oil. No matter if you have a nuclear powered aircraft carrier or not.
The most damage China can do to western interests is that they start some local war, which we disapprove of, and so we end up sanctioning and embargoing them, which means we lose their production. The end of mass production from China. This is how we killed the USSR and a bunch of smaller wannabes. It's just geography. USA and Europe have the most productive land and port access. Most access to oil to give us long reach. China is practically land locked because of the first island chain. Think of it as being surrounded by a thousand aircraft carriers with infinite fuel. They don't have enough oil domestically. Middle eastern oil is trivially interdicted.
i doubt that - the USSR killed themselves with their poor economic policies (as well as expensive wars they cannot really afford).
China on the other hand, hasn't made the same mistakes. They posture, they build military installations (such as those in the south china sea), but they haven't committed much, if at all, to an actual war. Not to mention their economic policies are vastly superior to the USSR's - sucking in western capital at the beginning, and now, to overtaking them.
China's lack of oil is an achilles heel, but not a very big one. And every day, they're closer to diversifying away from oil as an energy source. Not to mention having pipelines through russia and the much of asia minor to have land routes that aren't blockade-able.
The economic modal of china is irrelevant to how their geography makes them easy to blockade. I think you might have just misread what I was saying here. I was saying how they don't have a viable long game _if_ they choose to start a war that the west doesn't approve of. Because they are practically land locked like the USSR when compared to the west.
Nothing replaces oil as an energy source for military reach.
China is decades away from having any pipelines to north siberia where most of the russian oil is. There's really nothing practical even on the drawing board atm.
Pipelines are the easiest transport of oil to stop. Look at ukraine.
They should set up a foundation/LLC if they don't have one already and require a support contract for fixing any bugs in niche codecs. Target the 95%-98% use cases for "free" work. If someone gives them a CVE on something ancient, just note it with an issue tracker and that the problem is currently unsponsored. Have default build flags to omit all the ancient/buggy stuff. If nobody is willing to pay to fix all the ancient crap, then nobody should be using it. But if someone is willing to pay, then it gets fixed.
I'm pretty well "on the spectrum" and people glazing me in real life produce suspicion and discomfort rather than any good feelings.
I don't have a problem just ignoring all the LLM glazing, although I'd really like the ability to turn it off.
The fact that they've all been trained to do it, because so many of the "normies" fall for it, is kind of an indictment in my eyes. Bit of a mirror held up to society.
You should probably be worried about how fake flattery works so well in society, and how this enables sociopaths and narcissists to flourish and control everything.
It seems we have a sort of Cantillon Effect in wages (spilling over from IT down into HVAC and food service), but not in goods.
And I'm not sure the AI/LLM focus sheds any light at all on the situation. Unfortunately, AI/LLMs aren't AGI and they're just a tool. They're just a better machine (assuming it all pans out) that radiologists all need to become experts in using. The way to analyze the situation isn't that the radiologists are the "bottleneck", the radiologists are the experts who are trained to use the tools.
reply