Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

DeepSeek uses banned Nvidia chips for AI model, report says

finance.yahoo.com

Post reply on HN
Print view View original

271–280 of 331 posts

Re: DeepSeek uses banned Nvidia chips for AI model, report says

#271

Post by InkCanon »

nostrademons wrote: post #119

I thought this was common knowledge. DeepSeek’s Wikipedia entry says that they trained all their models on Nvidia chips procured before the U.S. embargo to China on them. It wouldn’t surprise me if they continued acquiring them through, well, less than legal means. I also read somewhere (not Wikipedia) that they trained on ChatGPT, Claude, and Gemini queries, basically feeding in the output of competitor’s LLMs as tr…

It's not even stealing. They paid OpenAI for the tokens. It violates the OpenAI TOS, which specifically forbids using it's outputs for training competing models (which is very ironic)

Re: DeepSeek uses banned Nvidia chips for AI model, report says

#272

Post by rapind »

ignoramous wrote: post #183
voidnap wrote: post #158

Earlier quoted context omitted.

Yes. The bans are export controls. They are not banned in china. They are just banned from export in the US. Using them in china is legal in china.

> Yes. The bans are export controls. These export controls increasingly look like "tax". The White House said the US government would take a 25 percent cut of the chip’s sales, similar to a deal with AMD and Nvidia earlier this year that allowed them to sell lower-powered AI chips to China while paying the US government 15 percent of the proceeds. > Using them in China is legal in China. Technically, yes. The CCP, th…

> Technically, yes. The CCP, though, wants to incentivize Chinese firms to use domestically-manufactured chips.

This couldn’t be playing out better for Xi. Trump is China’s best president.

I used to think Trump was clueless and being outplayed, but now I realize he’s just looting and couldn’t care less about protectionism or the American worker.

Every single action from this administration can be explained by greed and ego.

Re: DeepSeek uses banned Nvidia chips for AI model, report says

#273

Post by mkl »

secret-noun wrote: post #174
jhfdbkofdchk wrote: post #72

Earlier quoted context omitted.

Do they all live at the same address of the overseas freight forwarder too? I've sold stuff on eBay to someone in Europe who had me ship to the same address in Delaware. I was confused so I googled the address and turned up the freight forwarding service.

This has happened to me a couple of times with eBay sales. Is it safe to transact with people who use freight forwarders in your experience? Do you lose any protections? Out of fear, in my cases, I cancelled the auctions. On second thought though, I wonder if it's actually the buyer using the service that is more at risk (introduction of 3rd party, more complex delivery, probably impossible to return, etc)

What's not safe for you? They pay you the money, then you send the item to the address they ask. You already got the money! Cancelling the sale because the buyer wants to spend a bit less on shipping seems like an awful thing to do. International shipping gets ridiculously expensive, so combining multiple small packages into one shipment makes perfect sense.

Re: DeepSeek uses banned Nvidia chips for AI model, report says

#274

Post by WhyNotHugo »

nostrademons wrote: post #119

I thought this was common knowledge. DeepSeek’s Wikipedia entry says that they trained all their models on Nvidia chips procured before the U.S. embargo to China on them. It wouldn’t surprise me if they continued acquiring them through, well, less than legal means. I also read somewhere (not Wikipedia) that they trained on ChatGPT, Claude, and Gemini queries, basically feeding in the output of competitor’s LLMs as tr…

> It wouldn’t surprise me if they continued acquiring them through, well, less than legal means.

Strictly speaking, it's not illegal for them to acquire it, it's illegal for an exporter in the US to sell (even if transitively) to them.

Re: DeepSeek uses banned Nvidia chips for AI model, report says

#275

Post by whimsicalism »

nsoonhui wrote: post #266
nostrademons wrote: post #119

I thought this was common knowledge. DeepSeek’s Wikipedia entry says that they trained all their models on Nvidia chips procured before the U.S. embargo to China on them. It wouldn’t surprise me if they continued acquiring them through, well, less than legal means. I also read somewhere (not Wikipedia) that they trained on ChatGPT, Claude, and Gemini queries, basically feeding in the output of competitor’s LLMs as tr…

I think the fact that DeepSeek trains on competitor queries (i.e., distillation) — along with using banned Nvidia chips — helps explain how it can achieve such low training costs (USD 6 million vs. billions) while delivering only slightly worse performance than its American counterparts. It also undermines the narrative that DeepSeek or China is posing a serious challenge to the U.S. lead in AI. The gap may be closin…

These models never cost billions to train and I doubt the final training run for models like GPT-4 cost more than 8 figures. 6 million is definitely cheaper and I would attribute that to distillation.

Re: DeepSeek uses banned Nvidia chips for AI model, report says

#276

Post by bluefirebrand »

rhines wrote: post #262
rllearneratwork wrote: post #260

Earlier quoted context omitted.

pretty sure it is against ToS for ChatGPT, Claude and Gemini.

ToS didn't stop the companies that built those models and it won't stop the companies that bootstrap off them. Until an AI company eats a multi billion dollar lawsuit for unlawful data use they will continue to operate this way.

> Until an AI company eats a multi billion dollar lawsuit for unlawful data use they will continue to operate this way

If only. That's my dream, massive copyright lawsuits against all of these AI players and maybe the courts can do something good for a change, put an end to all of this AI bullshit

Re: DeepSeek uses banned Nvidia chips for AI model, report says

#277

Post by nandomrumber »

rhines wrote: post #262
rllearneratwork wrote: post #260

Earlier quoted context omitted.

pretty sure it is against ToS for ChatGPT, Claude and Gemini.

ToS didn't stop the companies that built those models and it won't stop the companies that bootstrap off them. Until an AI company eats a multi billion dollar lawsuit for unlawful data use they will continue to operate this way.

Didn’t Anthropic already eat a $1.5 billion lawsuit?

Re: DeepSeek uses banned Nvidia chips for AI model, report says

#278

Post by whimsicalism »

int32_64 wrote: post #58

What's strange in this discussion of chips and export bans is there's been zero discussion of cloud access, I guess networked computers are difficult for America's gerontocrats to understand. I've rented H100s no problem on American servers and there's no KYC or anything, they let anybody do it.

sure but to do this at scale you need better control of the networking layer than they typically give you, and there will definitely be KYC at a certain point in the process

Re: DeepSeek uses banned Nvidia chips for AI model, report says

#279

Post by Aqua0 »

nostrademons wrote: post #119

I thought this was common knowledge. DeepSeek’s Wikipedia entry says that they trained all their models on Nvidia chips procured before the U.S. embargo to China on them. It wouldn’t surprise me if they continued acquiring them through, well, less than legal means. I also read somewhere (not Wikipedia) that they trained on ChatGPT, Claude, and Gemini queries, basically feeding in the output of competitor’s LLMs as tr…

"The TBD group is using several third-party models as part of the training process for Avocado, distilling from rival models including Google’s Gemma, OpenAI’s gpt-oss and Qwen, a model from the Chinese tech giant Alibaba Group Holding Ltd., the people said."

LOL. Distillation doesn't count as plagiarism, or you should call Meta out on it. They're distilling the Chinese model.

Ref: https://www.moneycontrol.com/news/business/inside-meta-s-piv...

Re: DeepSeek uses banned Nvidia chips for AI model, report says

#280

Post by Aunche »

int32_64 wrote: post #58

What's strange in this discussion of chips and export bans is there's been zero discussion of cloud access, I guess networked computers are difficult for America's gerontocrats to understand. I've rented H100s no problem on American servers and there's no KYC or anything, they let anybody do it.

OpenAi, Anthropic, etc have access to specialized network topologies from high profile relationships with cloud providers. I don't think you can get anything like that off the shelf.
Post reply on HN