China’s DeepSeek Surprise
https://www.theatlantic.com/technology/archive/2025/01/deepseek-china-ai/681481/?utm_source=reddit&utm_medium=social&utm_campaign=the-atlantic&utm_content=edit-promo
China’s DeepSeek Surprise
https://www.theatlantic.com/technology/archive/2025/01/deepseek-china-ai/681481/?utm_source=reddit&utm_medium=social&utm_campaign=the-atlantic&utm_content=edit-promo
29 Comments
Almost overnight, DeepSeek, the Chinese AI chatbot, has rocketed to popularity in the United States. Americans are divided over whether to embrace or fear it, Matteo Wong writes.
When the Chinese AI start-up behind DeepSeek-R1 launched its model, the program appeared to match the most powerful version of ChatGPT—and, at least according to its creator, had taken a fraction of the cost to build. The model has incited plenty of concern, Wong continues: “Ultrapowerful Chinese AI models are exactly what many leaders of American AI companies feared when they, and more recently President Donald Trump, have sounded alarms about a technological race between” the U.S. and China. But at the same time, many other Americans—including much of the tech industry—are lauding the program’s capabilities.
Unlike top American AI labs, which keep their research almost entirely under wraps, DeepSeek has made its program’s final code free to view, download, and modify—which means that anybody, anywhere, can use, adapt, and even improve upon the program. “That openness makes DeepSeek a boon for American start-ups and researchers—and an even bigger threat to the top U.S. companies, as well as the government’s national-security interests,” Wong continues.
The pressure is now on OpenAI, Google, and their competitors to maintain their edge. The release of this Chinese AI program has also shifted “the nature of any U.S.-China AI ‘arms race,’” Wong writes. With the relatively transparent publicly available version of DeepSeek, Chinese programs—rather than leading American ones—could become the global technological standard for AI. “Being democratic—in the sense of vesting power in software developers and users—is precisely what has made DeepSeek a success. If Chinese AI maintains its transparency and accessibility, despite emerging from an authoritarian regime whose citizens can’t even freely use the web, it is moving in exactly the opposite direction of where America’s tech industry is heading,” Wong continues.
Read more: [https://theatln.tc/E6ys7Mth](https://theatln.tc/E6ys7Mth)
— Grace Buono, audience and engagement editor, *The Atlantic*
My concern about an “AI race” between US and China is that nobody wants to stop and talk about ethics or safeguards. Any time you mention either the resounding response is that the other side isn’t observing ethics or safeguards and that we will fall behind if we do. All of this is, of course, in the name of pure profit at the expense of workers.
What surprise???
The models are public since mid December
How “strange” that this surfaces on or around the time that tic-tok is being scrutinized…..
This feels too good to be true. There’s going to be some alarming catch to this thing, or it’s not nearly as good as advertised or it’s actually just a copy of ChatGPT or something.
I just can’t trust a product that pushes CCP revisionist history, which bothers me much more than the made-up history I regularly get from the other models.
An incipient tech arms race between China and the US. Surely this will prompt Republicans to put funding in schools like during the Cold War, right? Right?
Is the surprise that the Chinese government is still an autocratic censorial shitshow?
Because that’s not a surprise.
How is the cost actually known or calculated for the Chinese?
We know they have a different economic model over there. I think maybe there are some steps being left out. Like government support etc.
Brace yourself for that rampant american nationalism that tolerates ZERO praise of China lol.
You can’t even talk facts about Deepseek on any sub related to tech or future tech bc everyone is drunk on the nationalism.
Citizens who cant freely browse the web.
Neither can us Americans. Our censorship is just as bad if not worse.
Well, so far the only thing the chat seems to be able to give me is a polite “read the fucking manual”, worded slightly different each time. Maybe it’s trying to tell me something, but I’m not getting it. Haven’t played with the API yet.
Isn’t this just basic free market forces at work? Shouldn’t we be happy for competition?
So this would make it cheaper to make my T-100s and revolutionise the robot slave industry. Cool cool cool.
I’m seeing a lot of mixed reviews on this program and I can’t sort the propoganda from the real positive reviews.
Either way, I will do everything I can to avoid giving China my data and money. ChatGPT does everything I need and I will not be switching over any time soon.
Deepseek feels like overhyped Chinese marketing and I really doubt it’s going to make any real, lasting ripples in the AI industry.
Everything is much cheaper when you can just steal the r and d from other companies.
So does it actually “match” the capabilities of Chat gpt and Claude and the like? Or is it a cheap Chinese knock off.
I also get that AI may be in the business of cheap knock off quality in which case it’s hard to tell which crappy AI is worse than its counterpart.
If you have the pro version you can use a us/eu hosted version of the deepseek model. Not a time defining discovery, but it’s an interesting data point imo
“There was no “surprise”. They put their focus on one thing, they’ve succeeded.
While America kept supporting wars, selling Military Industrial Complex weapons, China spent their money in something really productive.
What is interesting about this is that is uses the other expensive models to train the chinese cheap model which is just like China lol…. looks like the advantage this creates is that you can create cheap low budget models that are geared for specific applications and one that can be completely offline and run on cheap hardware.
So many missing the point.
It was done cheaply and does not require top tier talent or infrastructure. The threat here is not national, it’s existential to the concept of closed property. That this came from China is circumstantial, and the fundamental freakout is that the big bets made investors are all for a Rube Goldberg machine.
The emperor has no clothes.
Pure coincidence that Deepseek is quickly dropped when TT is being banished
Sounds like we are going to put more money into AI before that bubble inevitably bursts. Outside of checking grammar and finding coding errors, even the best ones are unreliable.
If I was standing in front of a tank piloted by Sam Altman or Elon Musk I’m confident they would gladly run me over.
I for one embrace our future techno communist overlords.
Feels like the same energy as the reaction of the US to the first Sputnik Probe
Is this the start of AI Winter II? The first AI Winter was the result of Expert Systems being extremely limited in terms of possible applications. So limited that there was no possibility of recouping the costs of development.
Now US companies have spent some hundreds of BILLIONS of dollars which means the industry needs to earn some tens of BILLIONS per year to achieve a decent ROI. That is turn a profit.
DeepSeek, having been made free to download means that companies like OpenAI will have a hard time monetizing ChatGPT. If it is sold on a subscription basis then competitors selling subscriptions to DeepSeek will have much lower costs simply by not having software costs to pay off.
I’m willing to bet the stole all the ip to make it happen. It’s also heavily censored and propaganda CCP.
Open AI moves to big profit model after saying how very important it is to maintain freedom and open source nature to avoid problems.
China startup takes the baton.
“Here world, here’s an open source version that is cheaper to build and far more powerful. Free. Enjoy.”
Becomes overnight success. Hugely popular.
Sam Altman and big corporations: big frowney face.
Open AI: *Shit. What have we done?*
*It’s too late. Your choices exposed you. Now you have to pay for it*
It may be trained on 6mil worth of data but they already had plenty of moneys worth in set up, it’s a very misleading statement.