Artificial intelligence keeps evolving and recently, a new AI app is out in China called DeepSeek is sending waves globally. This new AI may revolutionize the AI space and may be a great competitor to openAI’s ChatGPT. Here is how the new AI, DeepSeek works.
What is DeepSeek?
DeepSeek is a Chinese AI assistant that was created by a building startup, which was founded in 2023 by Liang Wenfeng. Liang is quite familiar with innovations regarding AI, considering he co-founded High-Flyer, one of those hedge funds predicting market movements by using AI. In principle, the technology behind DeepSeek is to be as strong as leading models from OpenAI or Google. However, here lies the twist: the development of this new AI and the operational costs are drastically reduced.
A lot of people in the U.S. have already turned their attention to this new AI because it recently soared to the top of Apple’s App store which shows that a lot of Americans are interested in the app. The app’s launch has led many to question assumptions about China’s ability to compete in the AI race, especially with restrictions on high-end hardware exports from the U.S.
How does DeepSeek work?
DeepSeek’s interface is straightforward and user-friendly. Much like ChatGPT, you can ask it questions or request help with tasks, and it responds conversationally. However, it does not generate images yet, unlike some other AI tools.
The AI model behind DeepSeek, called DeepSeek-R1, can process large datasets, recognize patterns, and provide thoughtful responses. Its training data is up to date as of July, but the app also includes a search function to fetch more recent information from the web.
One standout feature of DeepSeek is its efficiency. While most advanced AI tools rely on cutting-edge GPUs to operate, DeepSeek uses fewer and less advanced chips. This makes it cheaper to develop and run, offering accessibility to users without compromising quality.
Is DeepSeek safe to use?
Each time you use DeepSeek, it will collect data about your conversations, technical details about your device, internet connection, and more. While this itself is nothing out of the ordinary for AI applications, DeepSeek is a product of a Chinese company, so there are some very valid government access concerns.
DeepSeek also appears to respect the rigid internet rules of China. This involves the censorship of sensitive topics, such as the 1989 protests at Tiananmen Square. While these limitations might not be of concern to casual users, they could be an issue for users seeking total transparency and freedom with AI tools.
On the plus side, DeepSeek has opened the source for its AI models, allowing those outside China to modify the underlying technology to suit better security purposes and other free uses.
How did DeepSeek achieve so much with limited resources?
One of the most remarkable things about DeepSeek is its ability to compete with giants like OpenAI while using less expensive technology. According to a technical paper released by the company, DeepSeek trained its AI model using about 2,000 Nvidia H800 chips. These are less powerful versions of Nvidia’s flagship H100 GPUs, and they are now restricted from being exported to China due to U.S. regulations.
Training the model cost $5.6 million, a fraction of the billions spent by U.S. tech companies. This breakthrough has surprised industry experts and even caught the attention of investors like Marc Andreessen, who called it “one of the most amazing and impressive breakthroughs” he has seen.
What makes DeepSeek different from ChatGPT?
Although DeepSeek and ChatGPT have similar functions, they take very different approaches. Here is what sets them apart:
- Cost and efficiency: DeepSeek is built to run on less advanced hardware, reducing costs significantly. ChatGPT, on the other hand, relies on high-end GPUs, making it more expensive to operate.
- Accessibility: DeepSeek has made its AI models freely available for developers to download and customize. This open-source approach mirrors Meta’s strategy with Llama but contrasts with OpenAI, which keeps ChatGPT’s technology proprietary.
- Censorship and restrictions: DeepSeek’s adherence to Chinese regulations means it censors certain topics, whereas ChatGPT offers more open responses, depending on the user’s location and the service tier.
- Future development: While OpenAI plans to release more advanced models, DeepSeek’s efficiency and resourceful techniques might set a new standard for cost-effective AI development.
Even OpenAI CEO Sam Altman acknowledged the significance of DeepSeek’s model, calling it “impressive” but promising that OpenAI would deliver even better solutions.
Continue reading:
ChatGPT does not work: What to do in case the platform crashes and what are the AI alternatives?
The biggest ever fall in a company’s value – What happened to Nvidia’s share price?
