- Deepseek is a Chinese company AI, whose newest chatbot shocked the technology industry.
- Deepseek says his model rivals he’s main competitors, such as the O1 of Chatgpt, with a cost.
- Deepseek’s growth has influenced technology shares and led to the consideration of Big Tech’s massive investment.
An artificial intelligence company Based on China has shocked the industry of him, Sending some shares of American technology by plunging and raising Question if the United States leadership in it has evaporated.
Chinese start, Deepseek, revealed a new model of it last week that the company says is significantly cheaper to run than leading alternatives from leading US technology companies such as Openai, Google and Meta.
Here’s all you need to know about the new hot company.
What is Deepseek?
Deepseek is a Chinese beginning of artificial intelligence founded in 2023.
It has been the technology industry conversation after discovering a new flag model it last week called R1 on January 20 with a reasoning capacity that Deepseek says it is comparable to the O1 O1 model, but to a cost.
Deepseek made available the latest version of his assistant in his mobile app last week – and since then grew to become the best free app in the App Store app, exiting chatgpt.
Who is after Deepseek?
Deepseek began as a side project of Chinese entrepreneur Liang Wenfeng, who in 2015 established a quantitative protective fund called High Flyer he and algorithms used to calculate investments.
After buying thousands of Nvidia chips, Wenfeng began Deepseek in 2023 with high-fried funds.
That chatbot can be achieved using a free web account, mobile app or API.
Why are investors worrying about Deepseek?
Deepseek’s R1 model is built into its basic V3 model. The company said the V3 model was trained in about 2,000 Nvidia H800 chips at a total cost of approximately $ 5.6 million.
And although training costs are just part of the equation, this is still part of what other senior enterprises are spending to develop their basic models of it. Mark Zuckerberg, for example, announced that Meta plans to spend over $ 60 billion on capital expenditure this year as it doubles in him.
According to Bernstein analysts, the Deepseek model is estimated to be 20 to 40 times cheaper to execute than similar models from Openai.
The relatively low -declared cost of the latest Deepseek model – combined with its impressive ability – has raised questions about Silicon Valley’s strategy to invest billions in databases and infrastructure to train new models with the last chips.
Nvidia, a company that produces high -power potatoes essential for power models, saw its close shares Monday down nearly 17% on Monday, wiping hundreds of billions of its market cap. Other major technology companies have also been affected.
Deepseek also said that its models were mainly trained in less advanced, cheaper nvidia chips – and since Deepseek seems to perform as well as competition, which can pronounce bad news for Nvidia if giants Others of technology choose to reduce their confidence in the company’s most advanced chips.
What are Deepseek Technology Leaders say?
Deepseek’s success is also speaking to high -tech leaders.
Meta chief scientist, he, Yann Lecun, looked to endure some people’s panic for setting Deepseek in a post on the topic over the weekend.
Lecun said it is not as much as China’s advances are dancing ahead of JSC, it is more that “open source models are overcoming those of the owner”.
Director General of Microsoft Satya Nadella also weighed on X.
“Paradox jevons hits again!” Nadella posted on Monday morning, referring to the idea that innovation creates a request. “As it becomes more efficient and more accessible, we will see its use fall, turning it into a commodity that we just can’t get enough.”
Marc Andreessen, associate of Silicon Valley Andreessen Horowitz Entrepreneurship Firm, said in a social media post that “Deepseek R1 is the Sputnik moment”, referring to the Soviet Union satellite that shocked the US and assisted in the start of space race .
How does Deepseek compare with chatgt and what are its shortcomings?
Deepseek says its R1 model rivals O1 O1, the company’s reasoning model was discovered in September.
Like O1, R1 of Deepseek gets complex questions and disrupts them in more manageable tasks.
R1’s skill in mathematics, code and reasoning tasks is possible thanks to its use of “pure reinforcement learning”, a technique that allows a model to learn to make its own environment and incentive decisions.
Similar to Chatgpt, Deepseek’s R1 has a “Deepthink” way that shows users the reasoning or chain of machine thought after its production.
The Business Insider Tom Carter tested Deepseek R1 and revealed that he seemed capable of doing most of what Chatgt could. The app looks similar to that of the chatgpt, with a rare interface dominated by a text box.
However, one of the few things R1 is less capable is the answer to questions related to sensitive issues in China. For example, when Carter asked Deepseek about Taiwan’s status, Chatbot tried to direct the topic back into “mathematics, coding and logical problems, or suggested that Taiwan was an” integral part of China “for centuries.