Friday, February 14, 2025
HomeWhat is known about DeepSec, the new AI that outperforms ChatGpt.

What is known about DeepSec, the new AI that outperforms ChatGpt.

A new player is emerging on the global AI scene, and this time it’s not from the United States: it’s Deep Seca Chinese laboratory specializing in developing advanced open source language models.

DeepSeek R-1, a Chinese application based on generative artificial intelligence models, has climbed to the top of the App Store’s free download rankings in both China and the US in the past few hours, even surpassing the popular ChatGpt

How was it born? Deep Sec

Born in 2023 as part of Quantitative Fund. The high flyerDeepSec quickly gained global attention thanks to its innovative models. In particular, the launch of DeepSeek-R1 in January 2025 created a real buzz in the AI ​​community. The reason? DeepSec claims that R1 not only matches, but actually outperforms the OpenAI-o1 model on several key criteria, while operating at a significantly lower cost.

The startup says it has made significant progress in training models using significantly fewer Nvidia chips than U.S. rivals, fueling suspicions that Silicon Valley has over the past year. Billions of dollars are estimated to be invested in artificial intelligence.

The success of DeepSeek-R1 is based on a series of architectural innovations. is one of the important Multihead Latent Attention (MLA), a technique that reduces memory consumption by 40%, makes the model more efficient. Another distinctive feature is the architecture. A mix of experts (MoE), which allows Chinese artificial intelligence to activate only a fraction of its 671 billion parameters for each input, thereby optimizing computational resources.

How the company overcame US sanctions

A particularly interesting aspect of DeepSeek-R1’s development is how the team managed to overcome this. Restrictions China was imposed by the US government on the export of advanced semiconductors. HighFlyer’s decision to venture into AI is directly linked to these constraints. Long before the ban, founder Liang bought a large stock of Wenfeng Nvidia A100 chipone type is now banned from export to China. It is estimated that DeepSec has 10,000 to 50,000 units of these chips.

Recognizing the potential of these stocks for AI training led Liang to found DeepSeq, which was able to use them with low-power chips to develop its models.

What security risks?

Despite its impressive technical capabilities, DeepSeek-R1 also raises important data security questions. The Chinese raise concerns about possible issues with Wenfeng’s close ties to the Communist Party. technical espionage, The party is increasingly focusing on artificial intelligence research and development in the United States.

Additionally, DeepSeq’s ability to process and analyze large datasets in real time makes it a powerful tool for identifying vulnerabilities in complex systems. Unlike traditional cyber attacks, which rely on manually finding vulnerabilities in networks, software or infrastructure, DeepSec can automate the process at unprecedented speed and scale.

If not properly controlled, DeepSec could not only enhance China’s cyber capabilities, but also redefine global norms on privacy and security. data security, with long-term effects on democratic institutions and individual liberties.

RELATED ARTICLES

LEAVE A REPLY

Please enter your comment!
Please enter your name here

Most Popular

Recent Comments