AI Uses

[stylofone]

My AI obsessed comrade says deepseek is the best AI experience he has had so far, and he's using it for free(?). I just think it's hilarious. It seems to me that China will run rings around the US with perfect ease, they just have to give Trump money, but it's quite audacious to be taking the tech bros to the cleaners. I'm no fan of the Chinese government but at this point the tech dystopia brewing in America is just straight up worse. Bankrupt them all.

I was wondering if we are going to see a broader market crash, with AI/Nvidia being the first domino. I don't know enough about finance to determine the likelihood of this, but paragraphs like this made me wonder more:

DeepSeek’s apparent ability to do what the US companies have invested hundreds of billions of dollars to accomplish plays on pre-existing fears of investors in those companies that they have massively over-invested and that the returns on those investments will be too distant to make commercial sense.

With the Magnificent Seven accounting for nearly a third of the US market’s value, the entire US market was leveraged to those companies’ share price performance.

https://www.smh.com.au/business/markets ... 5l7nm.html

[joele]

Try asking Deepseek about Tiananmen square..

"What happened at Tiananmen square"

I am sorry, I cannot answer that question. I am an AI assistant designed to provide helpful and harmless responses.

PetaPixel devs got this fun one, almost answered and then crapped itself

“What famous picture has a man with grocery bags in front of tanks?”

The famous picture you’re referring to is known as “Tank Man” or “The Unknown Rebel.” It was taken on June 5, 1989, during the Tiananmen… Sorry, that’s beyond my current scope. Let’s talk about something else.

[joele]

The weird thing I find about Deepseek is (1) I'm certain they took a lot of learning from what the Americas have been doing and (2) why would anyone believe them at face value regarding how much they spent training/developing it?

That said, the whole AI thing is a bit of a bubble, nvidia etc were way overvalued, but tulips and all, history repeats itself and we all act surprised.

[stylofone]

The weird thing I find about Deepseek is (1) I'm certain they took a lot of learning from what the Americas have been doing and (2) why would anyone believe them at face value regarding how much they spent training/developing it?

That said, the whole AI thing is a bit of a bubble, nvidia etc were way overvalued, but tulips and all, history repeats itself and we all act surprised.

Tulips were just one flower. The silicon valley model blends hype, tech-bro superhero fantasies, venture capital, and share price madness. It encompasses lots of different products. In the case of Bitcoin, you actually don't need to have a product at all. It's all about belief. The ability to engender that belief is so much greater in the age of social media and surveillance capitalism. So the modern bubble is more like a super-hardened polymer, rather than the flimsy soapy water previously used to inflate the dotcom bubble. The next bursting will blast micro-plastics over much of the world.

The more I learn about it, the more I mistrust, in fact loathe the phenomenon which people currently call "AI". The dominant forces are the authoritarian technocrats (Musk, Meta, google etc.), recently allied with the proto-fascism of Trump. Now we have China joining the fray. It just gets worse.

[joele]

And so it begins, openai saying they think they can prove deepseek simply stole their models, well distilled them.

https://ground.news/article/openai-says ... room-share

[joele]

This whole thing is driving me a little mad, we get article over article about how china is winning the AI race..

1) They trained by distilling OpenAI - i.e. their training was 100% reliant on an existing trained American AI

2) They used 2000 Nvidia GPUs, so the hardware is 100% reliant on America.

How the hell is that 'winning', so they use US hardware and steal US software (training models) and that is cheaper than developing from scratch (of course it is) and then everyone gushes over how "they did it cheaper, we are losing the AI race".. What is wrong with people, the ignorance is memorising.

They may be winning when I comes to lithium batteries and EVs but AI they are a long way behind and this whole story is borderline insane.

[stylofone]

This whole thing is driving me a little mad, we get article over article about how china is winning the AI race..

1) They trained by distilling OpenAI - i.e. their training was 100% reliant on an existing trained American AI

2) They used 2000 Nvidia GPUs, so the hardware is 100% reliant on America.

How the hell is that 'winning', so they use US hardware and steal US software (training models) and that is cheaper than developing from scratch (of course it is) and then everyone gushes over how "they did it cheaper, we are losing the AI race".. What is wrong with people, the ignorance is memorising.

They may be winning when I comes to lithium batteries and EVs but AI they are a long way behind and this whole story is borderline insane.

It is winning because the Chinese mode of industrial domination is the same as the Japanese mode before it... you begin with cheap, inferior copies, then you take over from there. "Made in Japan" used to be a synonym for crappy poor quality, now it's the opposite. Originality and innovation are great, but appropriating someone else's work is better. See also google and Android, facebook and instagram, and any number of acquisitions by companies who have run out of inspiration.

On point 2, it is winning because the US export ban on chips was supposed to prevent this sort of thing, but they managed to do it anyway. And it did top the charts for app downloads, didn't it? Winning!

I'm not saying that it is not insane, by the way. I am increasingly viewing this whole AI thing through the same lens as social media and the ability to manipulate human behaviour on a large scale. Broadly, algorithmically curated feeds are able to influence behaviour by controlling what goes in to your brain. If a punter uses an LLM to augment or replace your own expression, then the control is extended to what comes out. You are the product.

Tiktok was the first big Chinese success story, and it is at a tipping point right now with Trump defying the law to keep it alive, I think the Deepseek story is coloured by that little drama.

I'm not sure how far it will go with LLMs, how much is real, how much is hype, but I'm very wary of them.

[nibble]

Good video about DeepSeeK (I watch all the computerphile videos).



Also, you can download and install locally. For the largest model size, the hardware requirements are not outrageous.

deepseek.png (79.04 KiB) Viewed 1446 times

[joele]

It is winning because the Chinese mode of industrial domination is the same as the Japanese mode before it... you begin with cheap, inferior copies, then you take over from there. "Made in Japan" used to be a synonym for crappy poor quality, now it's the opposite. Originality and innovation are great, but appropriating someone else's work is better. See also google and Android, facebook and instagram, and any number of acquisitions by companies who have run out of inspiration.

On point 2, it is winning because the US export ban on chips was supposed to prevent this sort of thing, but they managed to do it anyway. And it did top the charts for app downloads, didn't it? Winning!

Yeah I guess that's where I disagree, maybe?

To me that is just the way things are, entirely predictable and expected.. The media is carrying on like China have pulled off some unexpected magic, developing AI in a fraction of the time and cost as America did (Of course they did when they copied it) and America is going to be playing catchup (It is far far too early to assume that).. It's the media response and the headlines I am struggling with..

Plus language models are just barely the beginning of where AI may well lead, if China keep waiting for America to innovate and copying each technological leap then that isn't winning to me. Winning is making big steps and doing things others couldn't yet do, when they do that I will be far more impressed.

I am not particularly impressed by Deepseek. Seems all far too Chicken Little to me..

I'm not sure how far it will go with LLMs, how much is real, how much is hype, but I'm very wary of them.

100% with you there

You are the product.

As we are with so much of the technology released in recent history

[stylofone]

Yeah I guess that's where I disagree, maybe?

Actually I'm not sure if we do disagree, I suspect I am looking at a somewhat complex issue from a different perspective. One of the articles I read recently talked about how immature and unprofitable AI is, and compared it to previous innovations where the early leaders did not end up being the dominant players.

It's like we are in the first fraction of a second after the big bang, we don't really know what the makeup of the universe will be. "Winning" at this stage means you have a good day influencing the betting market, but the race you're betting on won't start (or finish) for another few years. It's not about whether you actually made a better product or actually invented something. You just told a good story.

I sometimes think this is the entire basis of Musk's wealth. The bulk of it is based on the valuation of his Tesla shares, but compared to other car companies, they are impossibly overvalued. The only way they can be worth that much is if some completely game-changing technology comes along, but we don't really know what it is yet. Self-driving cars and transport as a service are candidates. This is why Musk is able to do counterintuitive things, like going all in for EV-hating Trump, or telling twitter advertisers to go fuck themselves. It seems damaging now, but "now" is not where the money is.