doesn't threaten any models at all.
it threatens the money they get because it makes their asking price hyper inflated.
this post was submitted on 29 Jan 2025
66 points (97.1% liked)
Out of the loop
11298 readers
186 users here now
A community that helps people stay up to date with things going on.
founded 2 years ago
MODERATORS
It is an Chinese llm with the same focus as chatgpt. The main threat is that it claims to be cheaper to create and run. Less hardware needed and so on. Edit: it is also mostly open source so that is also a factor
On the last point, this reason it's 'mostly' open source is because it doesn't hit all the points from the OSI definition. Some news articles get that wrong when talking about DeepSeek, and other supposedly "open source" AIs
It's a model that was built with a way smaller budget and uses less computing resources than OpenAI's models. It's also open sourced (you can run it locally; or improvise it to your needs) unlike OpenAI's models.
The way AI usage is typically sold so it could be integrated to other tools is that you pay as much as you use. Deepseek's models significantly undercuts OpenAI's models in pricing while offering similar performance.
It is a an open source model that claims to have been trained for $5 million, while openai says they have spent 7 billion on research. So far it appears to run on par with chat gpt. This is causing investors to flock to competition to try creating the next one. Also, this causes investors to back off seeing a sudden bloom in cheap competition.
Deepseek is not open source, they just published their weights. If that's all it takes to earn trust, then people should also be fine using Meta's AI.
(People should not be fine using Meta's AI).
My apologies. I got fooled by the headlines. Looks like people are reproducing it anyway.
Nvidia has grown astronomically because they produce the big bad super GPUs that the big bad super LLMs allegedly need to be so big and bad and super.
Except we explicitly do not allow Nvidia to sell those products to CHINA. Because we've been trying to gain computational intelligence supremacy since like 1999, and we're coming to a tipping point.
However some small company just released a competitor (generally better) to the best big bad super LLMs out there. As a side project. And it is open source. Their budget was like .1% of any of the big bad super LLMs out there and they didn't use any of the big bad super GPUs (allegedly).
This makes Nvidias speculative astronomical growth over the last two years look hugely inflated.
Edit: China
Do you mean we don't sell nVidia GPUs to China?
Nope. Not the fanciest ones, anyways. The US has been cutting China off from high-end tech for years. In 2022, the government banned exports of NVIDIA’s A100 and H100 AI chips. When NVIDIA tried making toned-down versions (A800 and H800), the US banned those too in 2023. The goal was to stop China from developing advanced AI.
It’s not just hardware. Since 2018, the US has been restricting Chinese nationals from studying AI and robotics. In 2020, over a thousand visas were revoked to block researchers with ties to China’s military. On top of that, major US tech companies have been pulling R&D out of China, with IBM cutting a huge chunk of its operations in 2024.
People think China’s tech dominance is just happening on its own, but the US has been actively trying to choke it out.
But you said Japan earlier. I'm confused, but think it was just a typo?
Did you just confuse Japan (one of our closest allies) with China?
No, I didn’t confuse Japan with China. The US has been restricting China’s access to high-end AI hardware, research, and talent for years.
NVIDIA’s A100 and H100 chips were banned from export to China in 2022. When NVIDIA made weaker versions (A800 and H800), the US banned those too in 2023. The goal? Prevent China from developing competitive AI.
It’s not just hardware. Since 2018, the US has been limiting Chinese nationals in AI and robotics programs. In 2020, over a thousand Chinese researchers had their visas revoked because of “military ties.” US tech companies have also been pulling out of China—IBM, for example, slashed its R&D operations in 2024.
This has been a slow but deliberate strategy to control computational power. The fact that people are only just realizing it now says a lot.
But you mentioned Japan in your previous comment, and didn’t address that.
Except we explicitly do not allow Nvidia to sell those products to Japan
So... Yes you did.
We restrict those chips going to China. Not Japan.
You wrote a whole book confirming you misspelled the first time around without bothering to check what you had previously wrote. Odd choice
Threaten in the sense that it competes with, and is built under a different economic system. Got the free market folk all ruffled because it isn't playing the game the way they want it played.
It has it's own limitations of course, and the full strength version isn't quite 'run at home' but it is pretty close, and there are weaker versions that can even be run on an SBC (if slowly). It's definitely in reach of the enthusiast though.
It's also supposed to have some strong reasoning capabilities (haven't tested myself yet) which is a big thing compared to your common or garden chat bots that at best sometimes know facts.
Deepseek is build for censorship, ask it some stuff about tiananmen square
If your using the mobile app or a web app that runs your request on a server in china, yes, the responses you get will be based on their laws.
You can go get your own copy off of hugging face (assuming its not getting its own hug of death), run it locally and generate your own model/weights with what ever data you want and it will tell you anything you want. (You do have the option to use their pre-trained weights and you will get the same result as the first paragraph)
This is the reason Sam Altman has been crying himself to sleep for the last week, if you want an equivilent model from openAI, its in their most expensive plan (IIRC thats ~200$/mo)
It will actually talk about it if you run it locally
As if Western AIs didn't also censor things.
This one is open source. You can train over the omissions and run your own instance, unlike the big ones here
Look man, I'm all down for anti Chinese government sentiment but at least say something that isn't this fucking asinine
Have you tried Western LLM? It's so fucking censored it's basically useless.
chatGPT gives good answers on Tiananmen Square and also on Jan 6th 2021
OK child
Deepseek is yesterdays news. Now its all about Qwen. /s