Chinese start-ups such as DeepSeek are challenging global AI giants

[placeholder]

Their models are cheaper thanks to US export restrictions that have inadvertently spurred innovation

Last month, a Chinese start-up called DeepSeek astonished the international tech community with its latest open-source artificial intelligence model. DeepSeek-V3 delivers a performance comparable to that of better-funded US rivals such as OpenAI. This week it impressed once again with R1, its foray into AI reasoning.

https://archive.ph/51WIt#selection-2415.0-2419.1

 

I see lots of comments about how China just produces cheap copycat goods. Clearly this is an outdated view. Japanese manufacturing was regarded in the same way for about 25 years after WW2.

[John Drake]

And just in time, Musk has made Grok subscription only again. Was free last night but only gives a subscription prompt now. ChatGPT was already limited in the number of images it would generate for free and Gemini won't generate images of people unless you pay. They are driving people to DeepSeek, although no image function on DS yet.

[candide]

On 1/25/2025 at 4:39 PM, placeholder said:

Their models are cheaper thanks to US export restrictions that have inadvertently spurred innovation

Last month, a Chinese start-up called DeepSeek astonished the international tech community with its latest open-source artificial intelligence model. DeepSeek-V3 delivers a performance comparable to that of better-funded US rivals such as OpenAI. This week it impressed once again with R1, its foray into AI reasoning.

https://archive.ph/51WIt#selection-2415.0-2419.1

 

I see lots of comments about how China just produces cheap copycat goods. Clearly this is an outdated view. Japanese manufacturing was regarded in the same way for about 25 years after WW2.

That's smart from the Chinese to make it open source. By doing so they will be able to use freely any additional improvement made by anyone in the world. 

 

In the end the total amount of development resources dedicated to it may well exceed what is expected to be invested in ChatGPT and other LLM models.

[Walker88]

DeepSeek was developed for almost nothing (~$5.6 million), and apparently to maximize its value one won't need the fancy chips NVIDIA is making. Thus, NVIDIA lost $560 billion in market cap on 1/27/25. Also, the subscription models of ChatGPT and others were obviated, since anybody can choose open source DeepSeek for free, or close to it.

 

I'm beginning to think the $500 billion AI plan the President announced last week is just a thinly disguised Oligarch bailout, as the tech sector suddenly isn't worth the multiple the market gave it.

 

I wonder if tech insiders are going to be dumping even more? I wonder how bad margin calls are going to be for all the leveraged buyers of tech? I wonder what company Musk is actually going to run, or will he go full DOGE, as Twitter is worth very little compared to what he paid, $40-80,000 Teslas are in a market with $12-25,000 BYDs, and his rocket just blew up last week (2nd stage) and rained debris all over the Caribbean. 

 

I think we've seen lots of peaks in the last week....Peak Musk, Peak Equity Market, andalready Peak 47's Presidency. It's all downhill from here.

 

 

[stevenl]

They're doing it for a fraction of the costs the US companies are spending.

 

See the much more extensive post on this above mine 

[Cameroni]

On 1/25/2025 at 4:39 PM, placeholder said:

 

I see lots of comments about how China just produces cheap copycat goods. Clearly this is an outdated view.

 

Actually that's exactly what happened with DeepSeek. 

 

It may be the case it has managed to cut costs and compute, but we do know that it is built at least in part on the shoulders of the giants: it uses Nvidia chips – albeit older, cheaper versions - and utilises Meta's open-source Llama architecture, as well as AliBaba's equivalent Qwen.

 

https://www.bbc.com/news/articles/cx2jxvn0r51o

 

They copied Meta's open source. And by all accounts DeepSeek is not as good as the  US competition, for instance you can't ask it questions about Tiannamen Square. It censors itself when it comes to Chinese topics.