China’s DeepSeek

[shinnosuke]

Can anyone comment on this new juggernaut?

Sam Altman is apparently not happy.

[thecrazyone]

Someone tried to hack it today, lol

[shinnosuke]

Must have been Sam. It would be a shame if his $500 billion AI infrastructure grift got paid for out of his own pocket instead of via federal deficits.

Anyway, China did something great apparently with little fanfare and for pennies compared to Open AI, if I'm understanding this correctly.

[68Camaro]

Everything we know is 3rd hand at best, but either the progress is grossly overstated, or they are using illegal advanced US chips smuggled in quantities. They are using borrowed open source US origin software and borrowed chips (whether the lower performing legal ones or the ones they aren't supposed to have). Do I then believe the CCP on their results? The organization that lost control of the Wuhan virus? Not a chance. Everything is about appearance for them. The timing is also suspicious. But... it's probably good for the US innovators to think they might have some competition.

[pmbug]

I found two threads on X worth reading :

What were the innovations:
https://threadreaderapp.com/thread/1883 ... 07859.html

Why it threatens NVDI:
https://threadreaderapp.com/thread/1883 ... 95541.html

[68Camaro]

Thanks for the links. I don't follow all the techno-geek AI details, but in summary it sounds like they pulled in the boundaries of the self-teaching portion (which is where the time, expense, energy is) so that it gave an "80% solution" rather than a 99.9% solution. And it doesn't attempt to be all things to all people - it creates specialties that self-teach in one area, and are called on as needed. Sounds like the big guys are trying to create a God-like AI that knows everything about everything exactly, whereas they created a lowercase g god that knows about a lot of things, but not in as much detail, and it relies on "neighboring gods" to answer specialty questions. As those articles point out, nothing they did is unique, and can be implemented by anyone. The same things were bound to be done by someone else eventually, if they haven't already been and just hasn't been published. I think the only question is how it affects the GPU chip makers - is NVIDIA making chips that are more complicated than are needed, and /or not needed in as many numbers? Will innovations like this just cause people to go after more "intelligence" anyway.

[Recyclersteve]

It’s too early for me to really form an opinion on this.

That said, I do believe there are plenty of people out there who want the masses to panic buy and panic sell. This will create opportunities for people who manage money for others. It also creates opportunities for those who are comfortable selling short and trading options.

[shinnosuke]

https://www.marketwatch.com/story/why-d ... =home-page

The ramifications of Chinese startup DeepSeek, with its promise of delivering cheaper, more energy-efficient alternatives to harness artificial intelligence, have yet to be fully reflected in U.S. equities.

That’s the thinking of Don Townswick, director of equity strategies at Conning Asset Management, which oversees $170 billion in assets.

“If it turns out to be a little more sketchy, a little less valid than what people are saying now, the ‘Magnificent Seven’ stocks would tend to benefit,” Townswick told MarketWatch.

On the flip side, if DeepSeek ends up delivering a less costly way forward, “I think it’s going to be a lot easier for typical companies to more easily use AI in their business,” he said.

Under that scenario, Townswick sees benefits from DeepSeek that could be accretive to earnings for a broader mix of companies beyond the current AI heavyweights, through greater efficiencies and productivity from less-expensive AI solutions.