The AI application DeepSeek is making an unexpected splash in app store charts worldwide. But what is behind this Chinese AI, how does it perform in practice, and why does it censor certain sensitive information?
DeepSeek: A Rising Star in the Tech World
DeepSeek has recently become a hot topic in the tech industry. This Chinese artificial intelligence company has gained significant attention, but what exactly does it offer, and how reliable are its AI models? Let’s dive into the details.
What Is DeepSeek?
DeepSeek is a Chinese AI startup founded in 2023 by hedge fund manager Liang Wenfeng. The company specializes in developing advanced language models and artificial intelligence systems. DeepSeek is notable for leveraging open-source models, where the code is publicly accessible and can be adapted for various applications. According to the company, users’ personal data is stored on servers located in China.
On January 20, the company introduced “DeepSeek R1,” an AI-powered chatbot. The app has surged in popularity, securing the top spot among the most downloaded iPhone apps in the United States, even surpassing ChatGPT. In Germany, DeepSeek is also trending among the most downloaded apps in the App Store. Users can interact with the AI, upload files, and use it for web research.
What Makes “DeepSeek R1” Unique?
The R1 model takes a fresh approach to AI by breaking tasks into smaller steps, creating what is described as “chains of thought.” This allows the AI to simulate a reasoning process. Known as “Reasoning,” this method is similar to the approach employed by the U.S.-based OpenAI in a model it introduced in December.
DeepSeek claims its AI excels in areas such as mathematics and physics, potentially outperforming other chatbots like ChatGPT in these fields. What sets DeepSeek apart is its ability to articulate its reasoning before providing an answer to a query.
While U.S. investor Marc Andreessen has hailed the R1 model as “astonishing and impressive,” some tech experts caution against overhyping its capabilities. Analyst Ben Thompson, for example, described claims that R1 outperforms market leader OpenAI as exaggerated, although he agreed that the model is competitive.
Training and Language Performance
DeepSeek’s models have reportedly been trained primarily on English and Chinese text documents. As a result, prompts in these languages yield the best outcomes. The developers acknowledge that when users input prompts in other languages, the AI may still respond in English.
Conclusion
DeepSeek is making waves in the AI industry with its innovative R1 model, blending reasoning capabilities with advanced performance in certain fields. However, despite its rapid rise and strong reviews, experts advise a balanced perspective on its potential. With its focus on open-source development and global expansion, DeepSeek is a company to watch as it continues to evolve.