At the beginning of 2025, a new Chinese neural network DeepSeek appeared online, which experts are already predicting a bright future for.
The release of the new AI assistant was an unpleasant surprise for European conglomerates in the field of neural network creation: OpenAI, Apple, and Google. The fact is that its use is free.
Mixture of Experts Algorithm
Asian scientists have managed to develop a new learning model algorithm based on the mixture of experts architecture. Unlike standard algorithms that require expensive clusters and huge data centers, DeepSeek “learns” using the NVIDIA H800 resource. It accumulates expertise not immediately in everything, but in narrow areas, thereby reducing the training time from several years to several months.
The new Chinese neurochat has a wide range of functionalities, comparable to the globally renowned GPT. With DeepSeek, you can:
- generate programming codes, both simple and multi-component;
- search for information on the internet, where the AI prioritizes more current and fresh information and presents it not as links, like Google, but as a full-fledged detailed response to the query;
- translate images and PDF files into text;
- search for data in voluminous documents, with a fairly high accuracy rate.
But DeepSeek cannot draw pictures. However, another resource based on it, Janus Pro, can do that.
As of today, two working versions have been released:
DeepSeek-R1, which is more focused on programming and logical analysis.
DeepSeek-V3, which “likes” to work with large volumes of data and multiple languages.
Why is DeepSeek better than GPT?
- Completely free to use compared to $60 at GPT.
- More detailed analysis of large volumes of documents with precise search and systematization of necessary information.
- Ability to search for information on the internet.
- Open source.
Even the mere fact of the emergence of a new neural network in the AI field was such a big shock that OpenAI had to reduce the price of using its product by $10 to retain its audience.