DeepSeek V3 (neural network)

History

2025

Ban on employees of the US Congress apparatus

At the end of January 2025, it became known that American congressmen were banned from using the Chinese chat bot DeepSeek based on artificial intelligence. This measure, as noted, is related to security and administration issues.

According to the resource, Axios referring to a letter sent to departments of the House of Representatives of Congress, USA attackers use DeepSeek to spread malware software and infect various devices. In this regard, employees of the Congress apparatus are prohibited from installing DeepSeek on any official, smartphones computer and tablet.

𝓲

Фото: Freepik

DeepSeek is currently under review and is not currently authorized for formal use in the House, the notice said.

Startup DeepSeek from China, founded by 39-year-old Liang Wenfeng, caused a stir in Silicon Valley. The company has released the powerful DeepSeek V3, an open AI model that surpasses most counterparts, including ChatGPT. In addition, the startup has developed a free reasoning model R1 DeepSeek R1, which can compete with OpenAI o1. Against this background, the share price of many AI companies went down: investors around the world got rid of such securities in connection with the achievements of DeepSeek, which managed to turn into a competitor to the world's leading players in a short period of time.

Meanwhile, the American authorities began to study the question of whether the DeepSeek startup could acquire advanced AI accelerators Nvidia through third parties in Singapore in order to bypass US export restrictions. Even before the imposition of sanctions, DeepSeek purchased approximately 10,000 Nvidia A100 cards.^[1]

Mobile release, application supports Russian

In mid-January 2025, it became known that the Chinese company, DeepSeek which developed the eponymous based on chat boat , artificial intelligence released its mobile version. The application is available for operating systems and. Android iOS

DeepSeek has created the powerful DeepSeek V3, an open AI model that surpasses most open and closed counterparts, including. ChatGPT The model has 671 billion parameters. It is claimed that a data set of 14.8 trillion tokens was used in training DeepSeek V3. In programming-related tasks, the Chinese model bypassed the Meta Llama 3.1 neural network 405B (the company Meta is recognized as an extremist organization; activities in the territory of the Russian Federation are prohibited), OpenAI GPT-4o and Qwen Alibaba 2.5 72B.

𝓲

Фото: Freepik

DeepSeek-V3 demonstrates a breakthrough in the speed of infection compared to the solutions of the previous generation. This neural network tops the ranking of leaders among open source models and competes with the most advanced closed models in the world, says DeepSeek.

The mobile version of DeepSeek has wide functionality. The chatbot is able to answer questions on any topic, solve problems of various types, search for information on the Internet, compile a resume, generate program code, make excerpts from materials and much more. The DeepSeek app for smartphones provides the ability to create chats. In addition, the history of correspondence is saved for subsequent viewing. However, the voice communication function is not supported. The app is distributed free of charge through the official AppleApp Store and Google Play. You can download the chatbot, including from Russia - it supports the Russian language. Google Play^[2] link]

Notes