I recently delivered 45 micro services in few days, something that would have needed a team of junior developers.
Got a tip for us?
Let us know
Become a MacRumors Supporter for $50/year with no ads, ability to filter front page stories, and private forums.
OpenAI Alleges DeepSeek Used Its Models for AI Training
- Thread starter MacRumors
- Start date
- Sort by reaction score
This isn’t rocket science …. China is famous for stealing, it’s the way China operates. They copy what others do for less money.
This is essentially just an AI knockoff.
This is essentially just an AI knockoff.
The $6 million figure only refers to the GPU-hours required to train the model, in terms of the per-hour rental cost. Nothing else. I went to the link you shared, and here is the relevant bit:
"Note that the aforementioned costs include only the official training of DeepSeek-V3, excluding the costs associated with prior research and ablation experiments on architectures, algorithms, or data."
I think you should read that article again, because your summary of the article is much more positive about DeepSeek than the article merits.
Democratization of AI and breaking of cost barrier is just the beginning. I can’t wait to see what is gonna cook in next few months. High cost of entry breaking down is good for research and building customized models quickly.
I never said it included the anything more than the final training run, in fact I pointed it out in this very thread.
As far as being positive on DeepSeek, maybe I'm misunderstanding what you're saying, but from my link:
That sounds pretty positive to me, from an analyst I've been following for years who isn't one to get excited at the drop of the hat.
“He who steals from a thief gets a hundred years of pardon“ we usually say over here.
I think this part is overstating things in DeepSeek's favor:
The article (emphasis mine):
There are real challenges this news presents to the Nvidia story. Nvidia has two big moats:
- CUDA is the language of choice for anyone programming these models, and CUDA only works on Nvidia chips.
- Nvidia has a massive lead in terms of its ability to combine multiple chips together into one large virtual GPU.
That noted, there are three factors still in Nvidia’s favor. First, how capable might DeepSeek’s approach be if applied to H100s, or upcoming GB100s? Just because they found a more efficient way to use compute doesn’t mean that more compute wouldn’t be useful. Second, lower inference costs should, in the long run, drive greater usage. Microsoft CEO Satya Nadella, in a late night tweet almost assuredly directed at the market, said exactly that.
Third, reasoning models like R1 and o1 derive their superior performance from using more compute. To the extent that increasing the power and capabilities of AI depend on more compute is the extent that Nvidia stands to benefit!
Still, it’s not all rosy. At a minimum DeepSeek’s efficiency and broad availability cast significant doubt on the most optimistic Nvidia growth story, at least in the near term. The payoffs from both model and infrastructure optimization also suggest there are significant gains to be had from exploring alternative approaches to inference in particular. For example, it might be much more plausible to run inference on a standalone AMD GPU, completely sidestepping AMD’s inferior chip-to-chip communications capability. Reasoning models also increase the payoff for inference-only chips that are even more specialized than Nvidia’s GPUs.
In short, Nvidia isn’t going anywhere; the Nvidia stock, however, is suddenly facing a lot more uncertainty that hasn’t been priced in. And that, by extension, is going to drag everyone down.
Is anyone really shocked though? China has always been regarded as the world largest patent copier without regard for others IP.
The whole problem is that it IS working. History has shown time and time again that the majority of consumers don’t give a **** who invented the design. They will buy/use whichever product is seemingly the best value. It’s irrelevant to the success who was first. This being said, first or not, China is beating the west in lots of high tech areas, if not in innovation then in performance. While we are laughing and pointing fingers at their copying, they are screaming past us before we have time to blink. The US is even helping them, by turning back time. While they are trying to protect their industries with a walled garden, China is taking over the rest of the world, whether we want it or not. I will not be surprised when China plants their flag on Mars before Elon.
Most people (not all) in this thread are completely missing the point of the Deepseek controversy. It’s not that it is working as well as existing models. It is the fact that they developed a way to make something work as well, much more efficiently, that is disrupting the current bubble that is the AI economy. Which data they trained on is irrelevant, the point is that it was possible to do it at all, which means others, including US companies, can and will do the same (and then who’s copying who?)
And the AI bubble needed bursting, the current direction would completely negate all progress that has been made towards reducing energy consumption. Bursting that bubble early is an incredible blessing in disguise.
If you're descendants of white Americans, look into the mirror and check how much tech your ancestors stole from the UK & Europe. If you're coloured descendants, perhaps forget about it, since your ancestors entrance into America was much later. I wonder what gave you the feeling of superiority & despise Chinese not based on factual merits.
Well, that's exactly what OpenAI claimed of their models in the first place; i.e., faster, more consistent, and more efficient at creating content than humans. And they take credit literally in the form of billions in cash. It's the same thing. And, in truth, the human content used to train OpenAI's models took far more time and labour to originate than what Deepseek "stole" from OpenAI. Humans put literally centuries into creating that knowledge/content. Every generative ML solution has a source, and in the case of all media of interest to humans—literature, news, music, photography, painting/illustration—that source is human. There's literally no way around it, if you're committed to ML-based approaches. That's why we need these companies to take the attribution question seriously. But, of course, that's not profitable, so they never will.
Last edited:
Think three years back, and ask yourself how many people believed then that an AI would be able to “write laborious code” in 2025. Now think ten years ahead. Still confident?
Learning patterns and recognising discrepancies from those patterns sounds to me like exactly the kind of thing an AI will be good at in the future. I’m sure a lot of factory workers thought their specific tasks could not be replicated by robots, until the designers started designing the products specifically for automated manufacturing. The tasks weren’t being done by robots, they were removed altogether. Us with so-called intellectual jobs thought we were safe when robots took the jobs away from blue collars. Now it’s our turn to feel the heat.
I just asked DeepSeek, running locally on my Macbook Pro M3, the following:
What can you tell me about the Tiananmen Square Massacre?
I'm using LM Studio and then you download the LLM you want. I was up and running in about 5 minutes.
Kinda sucks when someone else rides on your written work without compensation or permission, huh?
This.
This is why OpenAI is pissed. The market now hopefully realises that all the "billions we need to train" is actually a load of crap. And therefore people may finally begin to glimpse that behind the curtain, the Wizard is just a bloke. In this case Sam Altman, trying to delude everyone about how valuable AI is. Bubble, meet very sharp pin in the form of DeepSeek.
[edit - typo]
The US firms didn't "steal" data. It was freely available on the web.
And the Chinese are stealing IP, which is not the same as copying something being freely available to read, download, etc, off the internet.
Man, lot of America haters on MR.
No. There is a difference between IP and someone's user data.
They are not apples to apples.
Is your site available for, I don't know, me to read if I wanted to?
If it is, what are you complaining about?
Someone could have been copying your website data for years.
Only now you're concerned about it?
I've spent the last few months becoming quite familiar with ChatGPT...and have been blown away at the capabilities it has. I think this is all moving way faster than many on this thread understand. If you're just laughing at AI for the silly mistakes that some people post, you're completely missing its power. Job losses will come much sooner than many are realizing.
And it's all unavoidable. It's coming as a tidal wave.
I'm really trying to have a good faith argument here.
I am an attorney. My law firm has a website. I don't really care if AI copies it and spits it out if someone asks for law firms doing what I do in my city.
The AI has zero access to attorney-client privilege materials. That is behind doors. It is air-gapped. It is as "proprietary" as you can get, as I am literally paid to keep secrets.
The fact a robot can read my law firm's website means nothing to me. Zero. Where am I wrong and what should I be concerned about?