Deepseek turned the world of AI on your head but still don’t fall for over

Ai just had your moment on Sputnik.

Entrepreneur Mark Andresen did this Bold claim on xThe social media platform known as Twitter last Sunday. The silicone valley, along with the stock exchange and online prognostics, all are spread by what seems to be a seismic level in the AI ​​space.

Deepseek Ai, a new AI model from China which is Jump to the top of Apple App StoreSends response to the Silicon Valley. Deepseek claims that AI is competing, and in some cases in some cases excelOpenai’s O1 reasoning model with some of the costs. Not only that, the Deepseek R1 model is a fully open code, which means that the code is open and anyone can use it for free.

A key differential between Deepseek-R1 and Openai of the O1 is that R1 allows you to see its chain of thought. This is an incredible idea of ​​how I “thinks”. You can actually see it Trying to answer questions about Tiananmanmen Square Before canceling his answer in the middle. Nvidia, the company that makes the chips supplying AI Revolution, saw it Stock immersed 18% and lost a Record $ 600 billion After climbing the Deepseek weekend. It makes sense. If what Deepseek says is true, it is an almost O1 performance for the apparently older NVIDIA chips while spending a small percentage of costs.

Commenting online are still trying to make sense of the sudden appearance of Deepseek on the AI ​​market. Does it actually perform with O1 at a lower price? To what extent can Deepseek and China’s claims be true to efficiency? Do the cost savings come from a great technical unlock or other areas in the supply chain in China make it more cheaper to use?

Nevertheless, R1 is impressive.

“This affordable price opens the door to the smaller companies and startup companies to use advanced AI technology, which was previously inaccessible,” says Mel Morris, CEO of The corpore you haveAI Research Engine, in a statement to CNET. Morris added that Depepeeek is a competition for created AI players and its “presence will probably stimulate faster advances in AI technology, leading to more efficient and affordable solutions to satisfy growing demand.”

Openai Executive Director may be implemented pricing For its MINI MINI requests near the end on Saturday.

As the Big Tech is constantly throwing billions of dollars, the processing of power and energy at AI, unlocking Deepseek efficiency can be similar to the type of jump we saw when the vehicles switched from carburetors to fuel injection systems. Unlike Openai, the Deepseek R1 model is open source, which means that anyone can use the technology. This is a major interruption on the market, currently dominated by Openai and Guell’s Gemini Chatgpt, both are closed and require users to pay to gain full access to their package of features.

In the AI ​​competition between the US and China, America has remained ahead thanks to the mass investment landfill of the Silicon Valley and the NVIDIA government blockade, which sells the largest AI chips to China. However, this blockade can only have stimulate China to make its chips fasterS Money, plus protectionism, is seen as a way to preserve China in second place, which makes the world rely on American technology. This dynamics may have shifted. Consumers and corporations around the world now have access to a highly executive model of “reasoning” of some of the costs. Not only that, the Mother Tiktok Bytedance company released Even a more expensive model up to r1S

As markets and social media respond to new developments outside China, it may be too early to say that America was beaten. But at least, China is quickly catching up.
“China has already produced GPT-4 quality models, but it has had a longer lag in time-as it took a year, a year and a half, something like that. But now there is a Chinese model that may be only six months ago and I think This is a difference, “said Lucas Hansen, co -founder of CivaiA non -profit purpose that uses software to demonstrate what AI is capable of. “So the US still has a lead, but not as big as before.”

One thing that will surely help AI companies catch up with Openai is the ability of R1 users to read their thought chain. Even if the R1 does not get every answer correctly, you can see how it can better develop it. The people of “Shock and Awe” feel with R1, comes from the ability to read their thought chain, according to Hansen. This is the Insight Openai has not provided access with its O1 model, as hiding the secret sauce supports people to fire monthly expenses for access to access.

However, there is a level of skepticism that must be taken with the price-performance ratio of R1. Thehe White Book that Deepseek has published has more than 100 co -authorsS This is a very brain force to train AI for a low cost of $ 5.5 million. This price of $ 5.5 million can simply be the energy cost of training the model, minus the individual salaries of every researchers, but China is not completely transparent about how it calculates these energy costs. The cost of creating a data center in China is probably different from the creation of one in the United States. And it is not certain whether the costs were subsidized by a cloud supplier or by the Chinese government itself, according to Hansen.

There is also skepticism of the Deepseek chips used to train your model. Is the company actually Using older chips NVIDIA A100 and H800 or is China has access to the most recent H100 chips through other meansAs Alexander Wang said, CEO of Scale AI.

Even if we accept this number of $ 5.5 million as a highly conservative estimate, it is still significantly smaller than $ 100 million is worth Openai for training on GPT-4The previous model of AI companies. Openai has not released numbers about what it costs to build O1, but given its much higher prices for tokens for customers, this is probably more expensive.

“With the load of the data center in the United States, it is expected to double or triple by 2030, any efficiency savings can have a significant impact,” says Mark James, temporarily acting in the Institute of Energy and Environment in the Law of the Law Vermont and graduation in a statement. Already, Utility services are highlighted by the high energy needs of AIS If Deepseek’s claims are correct, then this can greatly relieve the potential load of electricity, relieving stress on both users and environmental. “On the reverse side,” said James, “more effective models could unlock even more growth in the sector, which would mitigate the economy of efficiency and sharpen the stress on our network.”

Claims that the US has Lost AI war It may be premature. At least, the landscape immediately becomes more competitive and there is room to continue innovation. Deepseek also does not mean the world is on the abyss of achievement general intelligenceOr a super -advanced AI that is more in mind than people and can learn.

“I don’t think Deepseek approaches us one millimeter to the artificial general intelligence (AGI), but I think it brings us closer to commercial viable applications for a large language model (LLM), which is fantastic,” said applications, “said applications Ben Görzel Executive Director of the Alliance for Artificial Head (ASI) and founder of SingularityS Deepseek still has the same cognitive restrictions as other AI models. However, the effectiveness of Deepseek can democratize AI further.