Why DeepSeek’s AI Model Just Became the Top-Rated App in the U.S.

Why Deepsees has been the AI ​​model for the best US application

By Stephanie Pappas Edited Jeanna Bryner

Deepsees made a great wave of artificial intelligence Monday, the highest application in the Apple store and technology stocks are sent down. What is all the hustle?

Chinese start, Deepseeke, surprised the technology industry with a new model that engaged in skills OpenaiThe latest recent model – using lesser investment and low capacity chips. The US PRINTED PRINTED POINTS PRINTED TO CHINA EXPORTS AND SALE SALES OF CHINA. Deepseek, based on the eastern China of Hangzhou, had a high-performance Nvidia A100 chips, before the ban, were able to use his engineers to develop the model. But in progress, it says that the beginning was much smaller to train Nvidia H800 chips to train a new model to train DeEpseeek-R1 duplicate.

“So far we have seen how many successful technology companies that have worked, not necessarily the technology really,” says Ashlesha, the General Manager of the AI ​​Intelligence plane, Inc.

To help Science Journalism

If you enjoy this article, consider entering award-winning journalism Subscribe. By purchasing subscription, you are helping to ensure the future of stories about the discoveries and ideas that are conformed to today.

Alas in ordinary tests in mathematics and encoding, Deepseeek-R1 joined the AI ​​O1 open model scores Ventureeat. US companies do not introduce their large language models (LLMS) training cost, such as the systems that make known chatbots Cards. But Openai CEO Sam Altman In 2023 a spectator said Chatgt-4 workout cost more than $ 100 million. Deepseeek-R1 is free to download users, and a comparable version of the puppy costs $ 200 a month.

Deepsees does not have the cost of $ 6 million to reflect the cost of building a LLM from scratch, Naharaikar says; This cost can have the right tuning of this latest version. However, he said, better energy efficiency of the model would be more accessible to more people in more industries. The increase in efficiency could be good news in terms of AI script The impact, new data is a computing cost to create a LLM four and five times higher rather than typical search engine consultation.

Because it requires less computational power, the cost of runseek-R1 is the cost of similar competitors, says Hanchang Cao, an assistant professor in Hanchang Cao, a University of Information System and Operation Management. “For academic researchers or start-up, this cost difference means a lot,” Caok said.

Deepsees got his effectiveness in several ways, says Anil AnthaniyThe author Why Machines Learn: Elegant math behind the modern AI. The model has 670 trillion parameters, or variables Learns in trainingThe largest open source language is still explained by Anantashaswamy. The model uses an architecture called “Expert Mix”, so it is only an important division of these parameters to be more than several hundred billion trillion. This reduces computing costs. Deepseee also uses a method that LLM puts multi-heading attention; Instead of answering a word word, it generates many words at the same time.

The model is more among others like O1, how it strengthens how to learn in training. Many llms, while criticizing “criticism”, directing the errors, llm verified responses, Deepseeek-R1 uses the internal set of model rules. “Deepsee facilitated this process,” Anasthaswamy says

Another important aspect of Deepseek-r1 is that the company has done the code behind the open product products, as Anasthaswamy says. (Training data remains proprietary.) This means that the company’s claims can be verified. If the model Deepsee is as computational as the claims, he says it will probably open new avenues for researchers who use AI so quickly and cheaper. LLMS will also allow further research on their internal work.

“One of the big things has spread between the Academy and the Industry, the academy could not work with very large models or do research significantly,” Anasthaswamy said. “But something like that, it is available now for the academy because you have the code.”