Physical Address
304 North Cardinal St.
Dorchester Center, MA 02124
DeepSseek went to viral.
China AI Lab DeepSEeak entered into the next key consciousness this week Her ChatBot application rose to the Apple App Store Charts (and Google Play too). Calculated AI models, using the report effective techniques There are LED Wall Street analysts – and technologists – To ask the United States to protect the leadership in the AI race and if the AI chips are required.
How did Deepseek come from, and how did it react to international fame so quickly?
Deepseek’s trader Origins
DeepSseek is supported by the management of high flyer capital, which is a Chinese quantitative hedge fund, which uses AI to inform trading decisions.
AI enthusiasm Liang Wenfeng High flyer high flyer in 2015.
In 2023, High Flyer began to Deepseek as a laboratory dedicated to investigating AI tools separately from financial work. As one of the high flyers, Laboratory DeepSeek also entered its company.
From the first day, DeepSEEK has set up its own data center groups for model training. But like other AI companies in China, DeepSEEK was affected by export prohibitions in hardware. The company was forced to use the company’s NVIDIA H800 chips, a chip’s less powerful version of the H100, H100 companies.
Deepseek’s technical staff is said to say young people. Company Reported aggressively employed Ai researcher from the best Chinese universities. DeepSseek also hires people without any computer science Tech, according to the New York Times, to help you better understand better topics to technology.
Deepseek’s powerful models
DeepSeek first models – DeepSeek encoder, DeepSeek LLM and DeepSeek LLM – in November 2023, when the next-geneepseek-v2 family of the AI industry began to warn.
DeepSeek-V2, a general purpose text and image analysis system, is well played in various AI criteria – and was cheaper than comparable models in time. Deepseek’s local competition, including motive and sonsbiaba, had to reduce the price of some models and make others completely free.
Deepseek-v3In December 2024, it was only added to Deepseek’s infamy.
Deepseek’s internal benchmark test, DeepSEEK V3, both loaded and open models such as meta Llama And only API, “Closed” models that can be achieved via an API such as Openai GPT-4O.
Equally impressive, Deepseek’s R1 is a “justification” model. Released in January, the Deep Alms R1 is performing Openai’s O1 model in key criteria.
Being a substantial model, R1 checks an effective fact-itself, it helps to prevent some traps that normally browse models. Approximately the models take a little longer – usually up to a few minutes – to come to solutions compared to a non-typical model. The side is that they tend to be more reliable in domains such as physics, science and math.
R1, DeepSeek V3 and other models of DeepSeek, other models have a negativity. China becomes an advanced EU, subordinate calister To ensure the “embody the main socialist values” of China’s Internet regulator. DeepSeek will not answer questions about R1 Tiananmen Square or Taiwan Autonomy, for example, for example.
A disruptive approach
If DeepSeek is a work model, it is not clear what this model is. The company comes from the market value of the product and services for free for low prices and others.
Deepseek’s route, the progress of efficiency, allowed to maintain the competitiveness of the expenses. Some experts argue However, the figures presented by the company.
Whatever the case, the developers, the statements that allow commercial use, they are usually understood, they appealed to Deepseek’s models, which are commonly understood. Clem Delanguue is one of the platforms hosting Hugging Face, DeepSeek models Developers in Hugging Face have created more than 500 “derivative” models of R1 2.5 million downloads are combined.
DeepSeek has been successful against larger and more opponents Described as “continuing up to AI” and “Excessive liar.” The success of the company was at least in charge of responsibility Causes NVIDIA stock price to decrease by 18% on Mondayand for and for disseminate public response Sam Altman from Openai CEO.
Microsoft Deepseek’s Azure AI announced that it is available in the serviceMicrosoft platform that brings together AI services for enterprises under a banner. When DeepSEEK asked the EU’s influence on EU’s spending in the first quarter, CEO Mark Zuckerberg said Spending the AI infrastructure will continue to be “Strategic Preference” For meta.
At the same time, Some companies prohibit DeepSEEKAnd so Countries and government.
Deepseek is not clear what the future can do. Improved models are given. But it seems like the US government Be careful of what it takes as a harmful external effect.
TechCrunch has a papers oriented newsletter! Register here To get in your inbox on your box every Wednesday.
This story was originally released on January 28, 2025 and continuously updated with more information.
