It’s commendable what DeepSeek is doing by open sourcing and showing the chain of thought.
This title is misleading though - where it tries to state that DeepSeek only focuses on research and not revenue.
1. DeepSeek is reported to be profitable. It earns revenue from its API.
2. DeepSeek was founded by Liang Wenfeng who cofounded the quantitative hedge fund High-Flyer in 2015. The AI work is used at the hedge fund to trade and earn revenue. Liang Wenfeng is transparent about this in interviews.
3. DeepSeek has strong backing from the government now that they have made a break through. DeepSeek has received support from Beijing[1].
Personally I dont see this as a bad thing. Its completely logical and smart for a government to support break through innovation. We should do the same in the US.
> This title is misleading though - where it tries to state that DeepSeek only focuses on research and not revenue.
IMO, the title is quite accurate and supported by the content of the article. The team is prioritizing AI research at the expense of short term profit.
If the title was "DeepSeek is solely a research effort" you would have a case.
Agreed, per Liang Wenfeng's public comments, his aim was and remains to inspire his fellow Chinese geeks to lead in innovation and not be satisfied at simply following the lead, and creating better versions, of foreign (read Western) mind products.
Lest we forget, this is an option because others did figure out the expensive way how most of the things DeepSeek relies upon work. You might even call it research.
Yes, but the real question is: Why are they able to do that? I understand we're debating nuances here, but my concern is about the overall impression the title gives. It positions DeepSeek as some kind of higher ideal, yet the article achieves this impression by deliberately overlooking key facts.
For example, why can Google afford to run Waymo, a self-driving car company? Is it because Google prioritizes self-driving cars and safety over profit?
No. It's because Google's core business—selling advertisements, monetizing personal data, and essentially profiting from surveillance—generates enormous amounts of money.
With all of this said. I am a fan of DeepSeek and the amount of openness they have.
"Focuses on" does not mean "exclusively undertakes".
Hence "I focus on enjoying life more than chasing money" does not imply "I don't do anything for money". It does not preclude me from engaging in paid activities. It simply means they are not the main priority.
So pointing out that I have earned money is not a contradiction of my claim.
The comment seems to be entirely ignorant of the basic semantics of "focus".
> DeepSeek has strong backing from the government now that they have made a break through
This is one thing I'd like western governments to implement as well. When such a potentially world changing technology comes up, fund its research and socialize its benefits. Instead, all we take from China is more surveillance, censorship, and anti-encryption laws (UK)
Yeah the research result and IP should belong to the public commons, if the initial researchers get a head start with exploiting the IP im fine with that.
They are. There's a ton of research taking place at public institutions and by government contract. We socialize the benefit by charging taxes on the profits.
If you instead mean something like "we should give already rich billionaires even more billions to toy around with" then no. We should not do that. Let's cut the parasitic billionaires out, and just fund the researchers. You know, like we are doing.
you can complain about "parasitic billionaires", and some probably qualify, but academic institutions aren't majority-responsible for LLMs. transformers came out of google and, although some academics toyed with them, most of the interesting work has been from companies. i care more about results than sticking it to the "evil billionaire" or whatever.
perhaps the better question is why "parasitic billionaires" and their companies are producing better results than academic institutions, and how we can fix said institutions to get them running well again.
Yes they do. There have been broken by neoliberal policies for quite some time. Rich have gutted them out to force researchers into selling out to the private sector instead of just giving the fruits of the research to everyone under equal conditions.
entirely aside from where value accrues relative to where it's created, it's just true that universities aren't building what are pretty interesting and promising new technologies. you're mistaking a positive statement for a normative one, which is a depressingly common turn of events on the internet. it is simply true that all this stuff is getting built by corporations, which is not a value judgement as to that being good. as i said, we should be thinking about why corporations are beating out our academic institutions, because that's the starting point to fix them.
Opinions can vary on what is "interesting". Personally, I find LLMs extremely boring from a scientific standpoint. If they turn out to be a product, they could be extremely valuable, but if they fail in the marketplace, they don't offer any interesting insights into the nature of the universe of computer science. LLMs are a technology of engineering, not science.
What is clear though is that none of that "interesting" AI stuff would exist without the mountains of research done by academic institutions. Google may have cemented "Attention" as a concept in LLM style AI, but the math to do that didn't come from nowhere. That math isn't thought of as a "technology" because it's not valuable without some product development. Academic institutions don't do that product development. That's precisely what makes you think they're being "beaten"
OpenAI and DeepSeek are not beating academic institutions in any metric the academic institutions should care about. That's just true.
I actually see nothing from Liang suggesting that the work is motivated by or relevant for trading. Care to share your quotes?
What I do see though are these quotes that mildly contradict your POV:
“If we have to find a commercial reason, we probably can’t, because it’s not profitable.”
“We’re going to do AGI. It’s driven by curiosity.”
“Giants have users, but their cash cows shackle them”
“It’s like someone buying a piano for a home—first, they can afford it, and second, such a group of people are eager to play beautiful music on it.”
“Giving back is an honor, and it attracts talent.”
“We won’t go closed-source. We believe that establishing a robust technology ecosystem matters more.”
> This title is misleading though - where it tries to state that DeepSeek only focuses on research and not revenue.
That's not what the title says though. It says "research over revenue" not only research no revenue.
I found the title misleading for a different reason: the "unlike silicon valley" part. IMHO SV couldn't care less about revenue than innovation, experimentation and ultimately valuation. Stock prices have been long detached from revenue multiples.
Are DeepSeek's publicly models themselves used in the trading operations of the HFT? I'd assume not, I feel even the architecture of models must be significantly different when comparing what R1 does versus what I'd assume trading models would do.
Chinese AI treasure Deepseek has reportedly restricted some of its developers from leaving the country by requiring them to surrender their passports, likely due to concerns over data security and potential acquisitions.
Having worked in academic research, industrial research, and programming in the US, I agree strongly with the sentiment here. From what I've seen, many US companies--even those claiming to be research focused--don't really engage in actual research and confuse it with development. This conflation of research with product-driven iteration has created a gap left to be filled by US competitors
A company can't operate without a source of income. Even a non-profit still needs a source to keep the lights on.
In the case of DeepSeek, I believe it is being used as a tool by the CCP to collect data, propaganda and as an economic weapon in hindering western AI projects. So DeepSeek can make this claim if they are being secretly subsidised by the CCP.
This wisdom got superceded by "you are the product". It was an echo of the old days, now everyone and their dog phones your data home in some form and sells it.
I guess I got the meaning of cedere wrong. If it has this “to give way” sense, then yep, adding super to it makes no sense in this context. That said, you can add “super” to almost anything, since it means “over”. It just has to have reasonable meaning.
Ah fair enough. All words are invented and language is fluid and all that. I was just trying to be helpful because we all have our spelling blind spots.
While it would be nice to have some open source and research oriented players, it could also be a massively sponsored Chinese government effort to get people both domestic and foreign to spill their secrets to an entity from the PRC.
Keeping the model open has led to large number of Chinese companies to adopt it, which further increases the value as a spying tool.
I'm in Europe. With the current US administration, I could say the exact same thing about American companies. The US government is clearly willing to use every single commercial enterprise located in the US as leverage.
At this point China seems like the less harmful of the two choices.
The way I look at it, I would probably use China-produced models for EU or US matters, and a West-produced models for China matters.
Certainly, one would also combine the two and have a fully open source model (including the training data) with no censoring summarise the two outputs.
The enemy of your enemy is just your enemy's enemy.
Why does everyone insist on picking a side rather then having any pride in their own nations? Europe is large, educated and capable of it's developments. No one needs to pick anything.
> The enemy of your enemy is just your enemy's enemy.
That's a little much. I'm not at war here. Where traditionally the US has been viewed as a mentor, or maybe a benevolent parent, by and of the European union nations, It has now switched into that of an abusive boss. I'm not at war with my boss. He is not my enemy. I do however have to continually adjust my stance towards him, always taking into account that he would sell me in a heartbeat if there was a benefit to him. That the company is not my friend. That stops me from sometimes doing the most mutually beneficial thing, because I have to protect myself by taking a defensive stance.
I'm very hopeful for Europe, and thankful that my system roughly works. It's not like I loved the US policy before Trump, but now flipped. What is happening in the US was always what I thought was most likely, corporatism run amok. That doesn't make the US "an enemy" but rather an unreliable friend.
Why would anyone see these countries as being one another's enemy, though?
They are battling to be the leading superpower in the world, but they are not enemies: annual trade volume between them certainly attests to that.
In that sense, you can also look to extract value from either side. In the US, I'd definitely turn to China's depictions of US events to understand how it looks from another superpower's angle: you can still take that critically and not at face value, but it might be refreshing in that it's not polarised in the same way (Republicans vs Democrats).
Why does everything have to be seen through the lens of conflict and warmongering? The goal should be the betterment of all humanity, and an "adversary" benefiting from a global improvement in quality of life should not be reason enough to deny everyone breakthroughs in technology.
It doesn't, that's the joke. Yet pushing FUD has become the norm when an unfriendly country creates a clearly superior product.
Just like the whole TikTok ban -- TikTok didn't have any additional access or privileges than any other smartphone app, and used the exact same telemetry/analytics as FB, Instagram, etc.. and yet, the spyware/malware angle was pushed heavily.
The models aren't executable binaries. They are sets of network weights that can be run using various (typically open source) third party implementations.
Sibling comment already explains the open weights part; even if there was some small binary wrapper involved, it’s trivial to stop if from phoning home.
Approximately 7% of Chinese people are registered CCP members. CCP mandates that if a company has three or more CCP members, a Party branch must be established. Statistically, if DeepSeek has more than 100 employees, there is over a 97% probability that a CCP branch exists within the company.
Its happening for a while now. China is a huge contributer in various AI opensource projects. From video to audio to text. A very good thing and the reason why some of the US ai companies are in a somewhat desperate release mode. The only further thing we need is a strong hardware maker in China to bring back Nvidia in serious attention mode.
Research IS future revenue. The total failure of current industry leaders to remember their history of taking pure R&D, largely from publicly funded research, and then put into applied research and development tracks, was (could be is, but we're dismantling all that) something that heavily influenced the post WW2 economic boom and it's aftershocks.
The EU and APAC nations who want to bootstrap a fold forward will take advantage of this.
Isn't the big complaint that Silicon Valley doesn't focus on revenue. Like there was a pretty funny bit of the "Silicon Valley" show that was mocking this fact.
I would say for most use cases it doesnt matter if I can talk withit about what happend somewhere in china more than 30 years ago.
Also the deepseek-R1 is pretty low censored. It was shown that some of the censorship is in the frontend. Here on hackernews was an article for example where someone talked with the model in hexadecimal ascii numbers (if i remember correct) and it happily talked about tianmen etc...
The model is openly distributed, enabling others to improve it. Currently there are numerous uncensored and improved versions of Deepseek r1 out there already.
This community effect of improving models is why in my opinion the best LLMs will soon all be open.
So true... The Democrats would be a right wing political party in most European countries... The Republicans on the other hand are closer to extreme-right parties such as the Rassemblement National in France or the Aternative für Deutchland in Germany. By the way, the only leader that Vance met with was the leader of the AfD in Germany, which is pretty telling of their political orientation... In the UK, Elon Musk, for instance, is closer to Farage than to any other English leaders, which again shows an extreme-right trope from the current Republican administration.
What people in general don't realize is that "left" and "right" are nonsense words with no coherent relationship to modern politics. For example, are dictatorial totalitarians the right or the left? Notice that this category includes both Hitler and Stalin.
More retirees vote for Republicans than Democrats and then Republicans tend to do things like expand Medicare (e.g. Medicare Part D). Does that mean Republicans are on the left? They also seem to have few qualms about interfering in the market when it suits them, e.g. to subsidize oil companies with tax dollars. Is belief in free markets the left wing position or the right wing position? If it's the right wing position, doesn't that mean subsidizing oil companies with tax dollars is a left-wing activity?
You can define "left" as Democrats and "right" as Republicans, but then the US tautologically has both a left party and a right party. Or you can try to define "left" and "right" in some independent way, in which case the US doesn't have a left party or a right party because it basically has two kleptocrat parties each representing a different set of bandits.
Oh, and then he quote mines me ... I didn't state that most of this is wrong, I stated that people realize that most of it is wrong.
I have no obligation to specify exactly which parts of a series of baseless, illogical, and counterfactual assertions--plainly patent nonsense--I find to be wrong. "left" and "right" are not nonsense words; take a few political science courses, and perhaps some linguistics as well. The existence of an Overton Window does not make the terms "nonsense". And the terms originally come from the wings of the French National Assembly which was a real world division with specific meaning that still applies in a broadened sense.
Right, the analogous thing for people is "air". That's the first thing I mentioned. Breathing is not necessary in order to take notice of the air. It is sufficient, though.
But it's open with open research papers and auxiliary tooling etc. THE GOOD states can fine tune it with hatered, isolation and warmongering – they show how to do it in cost effective way.
