In terms of user experience, DeepSeek looks and works similarly to many other AI chatbots. However, it does not support voice or instant image generation like ChatGPT. The response speed is also inconsistent at times, especially during high traffic.
The practical experience of Lao Dong Newspaper's reporter shows that one of the remarkable features of DeepSeek is that the search engine can synthesize information from many sources on the web and give concise answers. This application also has the ability to identify and organize information accurately.
There is one major limitation, however: DeepSeek is only trained on data up to July 2024. So if users want to find the most up-to-date information, they will need to use the built-in search engine instead of relying solely on the AI model.
DeepSeek also supports summarizing documents from files uploaded by users or from images taken with phones. However, the accuracy of this feature is not always guaranteed.
In fact, an example given by The Washington Post shows that when scanning a photo of the Bay Bridge in San Francisco (USA), the application incorrectly identified the content on the image, thinking that there was text related to a Chinese website and the keyword "Xinjiang" instead of the actual information on the photo.
In terms of model modes, DeepSeek allows users to switch between different versions of the AI. The default version provides quick and detailed answers, while the new R1 model, called “DeepThink,” provides slower responses but clearly demonstrates the AI’s thinking process.
DeepSeek collects a variety of data from users, including chat content, device information, and “keystroke patterns.” This data is ultimately stored on servers located in China, similar to many other AI applications.
Overall, while DeepSeek offers a pretty good experience, it still has some limitations compared to larger AI platforms like ChatGPT. If you’re used to other AI applications, you may not see much reason to switch to DeepSeek at this point.
