Artificial Intelligence thread

[tokenanalyst]

The campaign to ban DeepSeek (and other Chinese AI labs) is gathering momentum.

View attachment 170168

I wouldn't be surprised if the only way to access DeepSeek will be via VPN a few years from now.

As for the accusations, the major US frontier labs engaged in massive copyright violations when they trained their models. This about pulling up the drawbridge under the guise of "national security".

Why is always overpaid Think Tank stooge with the humanities degree.

Doesn't matter. DS is open weight and free to download, there will always a provider even outside the US.
Yes, Chinese AI labs as we posted before here has been focusing in efficiency, especially DeepSeek, because the constant threat of US export controls. Efficiency is the biggest innovation of Chinese models in both training and inference.
They can cry me a river with the distillation crap. These clowns scrapped the entire internet and violated every copyright law in existence, including massive Chinese internet data and copyright works, without any compensation for the authors and companies. Also are you are going to tell me that Anthropic and OpenAI are not distilling data from Chinese models?

The biggest difference between Chinese AI labs and these AI grifters companies is that at least Chinese AI labs are giving something back.​

 

[tokenanalyst]

I said it before and I say it now. When the AI bubble pop, and will pop, I think the only thing left will be Chinese AI labs and probably google AI. I wouldn't even be surprised if even Nvidia goes down with the insane levels of debt of this mania. There is no way for Nvidia to offset the losses with their gaming business if the AI bubble pops, even TSMC will be in serious trouble given that a lot of their revenue is AI dependent.

 

[tamsen_ikard]

I said it before and I say it now. When the AI bubble pop, and will pop, I think the only thing left will be Chinese AI labs and probably google AI. I wouldn't even be surprised if even Nvidia goes down with the insane levels of debt of this mania. There is no way for Nvidia to offset the losses with their gaming business if the AI bubble pops, even TSMC will be in serious trouble given that a lot of their revenue is AI dependent.

Even if stock market value of these US companies craters by 80%, their dominance in their respective fields will still remain. Microsoft will still dominate cloud and windows, Meta will still dominate social media. Nvdia will still produce billions of dollars in AI chips. Yesz their revenue might go down, but AI will still only grow in use.

 

[tokenanalyst]

Even if stock market value of these US companies craters by 80%, their dominance in their respective fields will still remain. Microsoft will still dominate cloud and windows, Meta will still dominate social media. Nvdia will still produce billions of dollars in AI chips. Yesz their revenue might go down, but AI will still only grow in use.

You can't offset the loss of sale of a 1,000,000 dollar GPU with a 2000 dollar gaming GPU that only few gamers can afford. Google, Microsoft, Meta and Amazon have made massive bets on datacenters and AI related businesses. That is why they are making the best effort shoving AI down the throat of users even if the vast majority of users don't want to pay for it or don't need it. Microsoft is desperate, that is why they hired as CEO of their gaming division a person who knows more how to market AI products than videogames.

When hype dies down and businesses and investors find out that LLM models will be capable only produce slop code and slop text, diffusion models can only produce slop art. the whole thing will come crashing down. But don't get me wrong LLMs and diffusion models are useful but the insane hype that US AI companies had created around these models doesn't match reality.

If US companies marketed this technology as any other commercial product, like a OS or the smartphone, instead of hyping this models as super intelligent gods, the valuations and the build up would be more grounded in reality than the fantasy that we have today

 

[tamsen_ikard]

You can't offset the loss of sale of a 1,000,000 dollar GPU with a 2000 dollar gaming GPU that only few gamers can afford. Google, Microsoft, Meta and Amazon have made massive bets on datacenters and AI related businesses. That is why they are making the best effort shoving AI down the throat of users even if the vast majority of users don't want to pay for it or don't need it. Microsoft is desperate, that is why they hired as CEO of their gaming division a person who knows more how to market AI products than videogames.

When hype dies down and businesses and investors find out that LLM models will be capable only produce slop code and slop text, diffusion models can only produce slop art. the whole thing will come crashing down. But don't get me wrong LLMs and diffusion models are useful but the insane hype that US AI companies had created around these models doesn't match reality.

If US companies marketed this technology as any other commercial product, like a OS or the smartphone, instead of hyping this models as super intelligent gods, the valuations and the build up would be more grounded in reality than the fantasy that we have today

The current hype regarding AI is too much no doubt. Especially the agi talk. But that doesnt mean AI is not useful. They are useful if used correctly and their capabilities will only grow.

AI chip use will only increase and the biggest bottleneck will ultimately be power generation and grid stability.

Nvidia is not going back to gaming. GPU based mass processing will continue get ever more popular. LLM is just one use case. There are still plenty of usecases in Machine learning that needs mass gpu use. And LLM will not stop being useful, it will only improve in terms of architecture.

AI genie is not going back to the bottle. Your pessimism will ultimately prove incorrect.

 

[tokenanalyst]

The current hype regarding AI is too much no doubt. Especially the agi talk. But that doesnt mean AI is not useful. They are useful if used correctly and their capabilities will only grow.

AI chip use will only increase and the biggest bottleneck will ultimately be power generation and grid stability.

If the AI bubble pops and AI datacenter construction stop or even slow down, AI chip consumption will also slow down or alright stop, depending the severity of the slow down. In the 2000s when the dotcom bubble pop, at lot of telecom related companies died or almost died because they couldn't offset the losses when the internet hype slow down.

Nvidia is not going back to gaming. GPU based mass processing will continue get ever more popular. LLM is just one use case. There are still plenty of usecases in Machine learning that needs mass gpu use. And LLM will not stop being useful, it will only improve in terms of architecture.

Even if they go back to gaming that wont offset the losses of their AI related business.

AI genie is not going back to the bottle. Your pessimism will ultimately prove incorrect.

I never say AI will die out, but development would be more grounded in reality. When the 3D printer didn't replaced traditional manufacturing 3Dprinters didn't died but found their niche in the industry neither when the dotcom bubble the internet died.
I think LLMs and diffusions models are going to continue to exist after bubble pops but more grounded in reality and companies who hyped their valuation base on fantasy are going to suffer.
I maybe incorrect, who knows, but that is what looks like.

 

[klimsa]

AI chip use will only increase and the biggest bottleneck will ultimately be power generation and grid stability.

The biggest bottleneck is finance, of which power generation and grid stability are merely a part of, and in this regard, AI is on a 3 year timeline if we are being very generous. Yes, the technology will keep progressing, and will probably achieve some value add by obsoleting some low level jobs; the industry however is banking on straight up eliminating like 30% of all jobs total which is a far cry from the technology they actually have. They are making such outlandish promises I am actually starting to believe they know they are T minus 3 years to insolvency and are trying to implicate so much money in their game so as to force the fed to bail them out when it all goes down.

 

[tphuang]

Latest global open-source large model rankings: QWen 3.5 tops the list! Chinese-made models occupy eight of the top ten spots.

Please, Log in or Register to view URLs content!

Qwen has a lot of downloads and fine tuning, but it really is inferior to Kimi 2.5 and GLM-5 at this point.

 

[Bellum_Romanum]

Really am so very tired of Anthropic's whole "noooo if anyone develops AGI before us they're going to be the baddies!!! we need to be first!!" shtick

Doubly tiring when you consider that every single one of these companies know full well that language models do not/will not translate into AGI

According to Gen Z, the term "baddies" does not mean someone is terrible. Instead, "baddies" = hotties!

 

[Wrought]

Literally everyone is shitting on Anthropic, even the VC guys.

https://twitter.com/i/web/status/2026407209706692895