DeepSeek
Hangzhou DeepSeek Artificial Intelligence Basic Technology Research Co., Ltd. , commonly referred to as DeepSeek , (Chinese: 深度求索; pinyin: Shēndù Qiúsuǒ ) is a Chinese artificial intelligence company specializing in the development of open-source large language models (LLMs) . Headquartered in Hangzhou , the company is owned and funded by Chinese hedge fund High-Flyer , whose co-founder, Liang Wenfeng, founded the company in 2023 and currently serves as CEO . [ 1 ]
| 杭州深度求索人工智能基础技术研究有限公司 | |
|---|---|
 | |
| Type | Private |
| Foundation | May 2023 |
| Thirst | Hangzhou , China |
| Founder | Liang Wenfeng |
| Employees | Less than 200 |
| Area of influence | Information technology Artificial intelligence |
| Official website | deepseek |
The DeepSeek-R1 model has outperformed other contemporary large-scale language models, such as OpenAI 's GPT-4 , in tests with models optimized for image processing and complex data analysis . [ 2 ] [ 3 ] It was trained at a significantly lower cost, $6 million, compared to $100 million for OpenAI's GPT-4 in 2023, and requires only one-tenth the computational power of an equivalent LLM. DeepSeek's AI models were developed amid US sanctions against India and China over Nvidia chips , which were aimed at restricting those two countries' ability to develop advanced AI systems. [ 4 ] [ 5 ] [ 6 ]
DeepSeek makes its generative AI algorithms, models, and training details available as open source, allowing its code to be freely accessed, used, modified, and adapted to create new projects. The company actively recruits young AI researchers from top Chinese universities and also hires professionals from fields outside of computer science , with the aim of diversifying the knowledge and capabilities of its models. [ 7 ] [ 8 ]
On November 2, 2023, DeepSeek unveiled its first DeepSeek Coder model, which was free for commercial use and completely open source. [ 9 ]
On November 29, 2023, DeepSeek released the DeepSeek LLM ( large language model ), which scaled up to 67 billion parameters. It was designed to compete with other LLMs available at the time, with performance approaching that of GPT-4 . However, it faced challenges in terms of computational efficiency and scalability. [ 9 ] A chat version of the model called DeepSeek Chat was also released. [ 10 ]
In May 2024, DeepSeek-V2 was launched. The Financial Times reported that it was cheaper than its peers, with a price tag of 2 RMB for every million tokens produced. The University of Waterloo ’s Tiger Lab leaderboard ranked DeepSeek-V2 seventh in its LLM rankings. [ 11 ]
In November 2024, DeepSeek R1-Lite-Preview was released, designed to excel at tasks requiring logical inference, mathematical reasoning, and real-time problem solving. DeepSeek claimed that it outperformed OpenAI o1 on tasks such as the American Invitational Mathematics Examination (AIME) and MATH. [ 12 ] However, The Wall Street Journal stated that when using 15 problems from the 2024 edition of AIME, OpenAI o1 arrived at solutions faster than DeepSeek R1-Lite-Preview. [ 13 ]
In December 2024, DeepSeek-V3 was released. It came with 671 billion parameters and was trained in about 55 days at a cost of $5.58 million, using significantly fewer resources compared to its peers. It was trained on a dataset of 14.8 trillion tokens. Benchmark tests showed that it outperformed Llama 3.1 and Qwen 2.5, while also matching GPT-4o and Claude 3.5 Sonnet. [ 14 ] [ 15 ] [ 16 ] [ 17 ] DeepSeek's optimization on limited resources highlighted the potential limits of US sanctions on China's AI development. [ 18 ] [ 14 ] The total cost to train the model was $5.58 million and took about two months to complete. [ 14 ]
On January 10, 2025, DeepSeek released its first chatbot app, based on the DeepSeek-R1 model for iOS and Android . [ 19 ] Its launch resulted in a trillion-dollar stock market crash , [ 20 ] particularly for American and European technology companies, including Nvidia , which fell by $600 billion in a single day, the largest stock market crash in history. [ 21 ]
Controversies
Censorship
Some sources have noted that the official version of the R1 API uses censorship mechanisms for topics deemed politically sensitive by the Chinese government. For example, the model refuses to answer questions about the 1989 Tiananmen Square protests and massacres , the persecution of Uyghurs , comparisons between Xi Jinping and Winnie the Pooh , or human rights in China . [ 22 ] [ 23 ] [ 24 ] The AI may initially generate a response, but then delete it shortly afterward and replace it with a message such as, "Sorry, that's beyond my current scope. Let's talk about something else." [ 23 ] The built-in censorship and restriction mechanisms can only be removed to a limited extent in the open-source version of the R1 model. If the "core socialist values" defined by Chinese internet regulators are touched upon or the political status of Taiwan is raised, discussions will be terminated. [ 25 ] When tested by NBC News , DeepSeek's R1 described Taiwan as "an inalienable part of China's territory" and stated, "We firmly oppose any form of 'Taiwan independence' separatist activity and are committed to achieving the complete reunification of the motherland through peaceful means." [ 26 ] In January 2025, Western researchers were able to trick DeepSeek into giving accurate answers to some of these topics by requesting in its response that it replace certain letters with similar-looking numbers. [ 24 ]
References
- ↑ «DeepSeek: cheaper chip, self-censorship, 'threat' to the US... see questions and answers about Chinese AI» . G1 . January 29, 2025 . Retrieved February 1, 2025
- ↑ Vincent, James (28 January 2025). «The DeepSeek panic reveals an AI world ready to blow» . The Guardian . ISSN 0261-3077 . Retrieved 1 February 2025.
- ↑ «آموزش ثبتنام در دیپسیک و دسترسی رایگان به مدل DeepSeek V3» . شهر بورس
- ↑ «After DeepSeek case, US must monitor Nvidia exports more strictly» . Valor Econômico . January 30, 2025 . Retrieved February 1, 2025
- ↑ Mallick, Subhrojit; Lohchab, Himanshi (16 January 2025). «Biden admin's cap on GPU exports may hit India's AI ambitions» . The Economic Times . ISSN 0013-0389 . Consulted on February 1, 2025
- ↑ "Nvidia investigation signals widening of US and China chip war | Computer Weekly" . ComputerWeekly.com . Retrieved February 1, 2025
- ↑ Metz, Cade (January 27, 2025). «What to Know About DeepSeek and How It Is Upending AI» . The New York Times (in English). ISSN 0362-4331 . Retrieved February 1, 2025.
- ↑ Metz, Cade; Tobin, Meaghan (January 23, 2025). «How Chinese AI Start-Up DeepSeek Is Competing With Silicon Valley Giants» . The New York Times (in English). ISSN 0362-4331 . Retrieved February 1, 2025.
- ↑a b Se, Ksenia (August 28, 2024).«Inside DeepSeek Models».TuringPost. Retrieved December 28, 2024. Archived from theoriginalon September 18, 2024
- ↑ Sharma, Shubham (December 1, 2023). «Meet DeepSeek Chat, China's latest ChatGPT rival with a 67B model» . VentureBeat . Retrieved December 28, 2024 . Archived from the original on December 23, 2024
- ↑ McMorrow, Ryan; Olcott, Eleanor (9 June 2024). «The Chinese quant fund-turned-AI pioneer» . Financial Times . Retrieved 28 December 2024 . Archived from the original on 17 July 2024
- ↑ Franzen, Carl (November 20, 2024). «DeepSeek's first reasoning model R1-Lite-Preview turns heads, beating OpenAI o1 performance» . VentureBeat . Retrieved December 28, 2024 . Archived from the original on November 22, 2024
- ↑ Huang, Raffaele (24 December 2024). «Don't Look Now, but China's AI Is Catching Up Fast» . The Wall Street Journal . Retrieved 28 December 2024 . Archived from the original on 27 December 2024
- ↑a b c Jiang, Ben; Perezi, Bien (January 1, 2025).«Meet DeepSeek: the Chinese start-up that is changing how AI models are trained».South China Morning Post(in English)
- ↑ Jiang, Ben (27 December 2024). "Chinese start-up DeepSeek's new AI model outperforms Meta, OpenAI products" . South China Morning Post . Retrieved 28 December 2024 . Archived from the original on 27 December 2024
- ↑ Sharma, Shubham (26 December 2024). «DeepSeek-V3, ultra-large open-source AI, outperforms Llama and Qwen on launch» . VentureBeat . Retrieved 28 December 2024 . Archived from the original on 27 December 2024
- ↑ Wiggers, Kyle (December 26, 2024). «DeepSeek's new AI model appears to be one of the best 'open' challengers yet» . TechCrunch
- ↑ Shilov, Anton (27 December 2024). «Chinese AI company's AI model breakthrough highlights limits of US sanctions» . Tom 's Hardware . Retrieved 28 December 2024 . Archived from the original on 28 December 2024
- ↑ «Release DeepSeek-R1 deepseek-ai/DeepSeek-R1@23807ce» . GitHub . Retrieved January 28, 2025
- ↑ "Tech companies lose $1 trillion in market value due to Chinese AI 'threat'" . G1 . January 27, 2025 . Retrieved January 28, 2025
- ↑ Saul, Derek. «Biggest Market Loss In History: Nvidia Stock Sheds Nearly $600 Billion As DeepSeek Shakes AI Darling» . Forbes . Retrieved January 28, 2025
- ↑ Field, Matthew; Titcomb, James (January 27, 2025). «Chinese AI has sparked a $1 trillion panic – and it doesn't care about free speech» . The Daily Telegraph (in English). ISSN 0307-1235 . Retrieved January 27, 2025.
- ↑a b Steinschaden, Jakob (January 27, 2025).«DeepSeek: This is what live censorship looks like in the Chinese AI chatbot».TrendingTopics. Retrieved January 27, 2025.
- ↑a b Lu, Donna (January 28, 2025).«We tried out DeepSeek. It worked well, until we asked it about Tiananmen Square and Taiwan». The Guardian (in English).ISSN 0261-3077. Retrieved January 30, 2025.
- ↑ «The Guardian view on a global AI race: geopolitics, innovation and the rise of chaos» . The Guardian (in English). 26 January 2025. ISSN 0261-3077 . Retrieved 27 January 2025.
- ↑ Yang, Angela; Cui, Jasmine (January 27, 2025). «Chinese AI DeepSeek jolts Silicon Valley, giving the AI race its 'Sputnik moment ' » . NBC News . Retrieved January 27, 2025.
External links
Comentários
Postar um comentário