☆ Yσɠƚԋσʂ ☆, yogthos@lemmy.ml

China’s CXMT just broke the global memory chip duopoly held by Samsung, SK Hynix and Micron. Their IPO caused a major global semiconductor stock rout and Korean markets halted trading because it’s basically propped up by these companies. And on top of that, there are now increasing worries about the whole AI bubble built on circular investments. If data center build out stops then the demand for chips and memory is going to collapse too.

looks like at this point, it’s Iran in the driving seat and escalating against the US

haha same

Honestly, I think the most reasonable approach is just to see what other people’s experience is like and which models are well regarded, then try them out and see which one is the best fit for what you’re doing. You might not even need the top performing one necessarily, and speed or lower resource usage might be a bigger factor.

lol might have to smuggle it directly from China 🤣

Yeah, the new policy is they’re not going to seek reunification, but I think that mostly implies that they see no reason to take over the south by force. If there was a collapse in the south, integrating it would mean removing US presence from their border. And that would be very valuable.

My prediction is that social collapse in the south will eventually lead to reunification.

What I’ve realized is that people internalize a narrative about how the world works. Things that fit that narrative are accepted, while those that contradict it are rejected. Facts and truth have little role to play here because the narrative is what matters. There’s an underlying thermodynamic explanation for this. We don’t hold ideas in isolation because each idea is connected to many other concepts in our minds. The older we get, the larger this network of knowledge becomes, and all connected concepts must be at least minimally compatible. So when a person encounters a fact that doesn’t fit, they must either rework an entire network of existing facts to accommodate it or simply discard it. The latter is strictly cheaper in terms of energy use for the brain, which is why new ideas that contradict existing beliefs are very easy to ignore. This dynamic changes only when there is a physical pressure, such as an economic downturn, on the person, forcing them to reevaluate their beliefs. Only when the narrative becomes so divorced from material reality that it is no longer tenable do people start becoming open to new ideas that challenge their existing worldview.

I expect this is gonna be far worse because the economy was way more diversified back in enron days.

Yup, and another side of it is that having these disclosures lets defenders of the system say that there is transparency, and and that’s why, despite the flaw, the system really does work overall. Sure, all these terrible things happen, but there is accountability, and disclosure, so really it’s all fine.

https://en.wikipedia.org/wiki/Limited_hangout

lol basically

Sure, a benchmark doesn’t capture all the subtleties and different use cases, but it does give a general idea of the capabilities of a model. Obviously, you have to run the model and see if it does what you need. But the chart isn’t really about the nuance, it’s showing how drastically the efficiency of the models has improved in just a year. The fact that we can even reasonably compare a model you can run on a desktop to one that needed a data center just a year ago is phenomenal.

It really is amazing when you present people with all the horrible stuff that’s been revealed, and they go oh yeah that all happened, but that was in the past. Like you have the same government with the same interests, same social and economic structures, but somehow think that the same type of society somehow produces different results today than it did previously. Just absolute brain worms.

and nobody ever learns anything in the process

I’m fairly optimistic that people will figure out how to optimize the models a lot further going forward. One obvious path is to try and separate the reasoning network from the trivia that gets baked into the model, and some work is being done in this area. If you could have a context free reasoning engine and then feed the facts it needs to know on the fly based on the context you’re running it in, then you could likely have a much smaller model that’s very capable.

Not sure what Moore’s law has to do with anything here to be honest. The models you can run locally on a consumer desktop can do real work, and their resource usage is no different from any other software like games that you’d run.

The article doesn’t really talk about this directly, but when you read it critically it becomes clear that capitalist corporate structures are ill equipped to deal with the negative effects of LLMs.

Yeah, I’m not arguing against efficiency on the software side. The way I read the article isn’t really that it advocates for brute force, but rather that we should be focusing on general solutions because they scale. The work on efficiency should focus on how to make search and statistical inference more efficient. Incidentally, we see a lot of that work happening in China right now precisely because they don’t have as much compute to work with.