Huawei introduces DeepSeek R1 to block politically sensitive topics: All the details

deepseek

Huawei has launched DeepSeek-R1-Safe, a more modern version of the DeepSeek large language AI model, claimed to be nearly 100% successful in preventing discussion of politically sensitive topics, in compliance with Chinese regulators. As reported by Reuters, the model has been developed in partnership with Zhejiang University, a leading institution and the alma mater of DeepSeek’s founder Liang Wenfeng.

DeepSeek R1: Key details

According to a post on Huawei’s official WeChat account, the company has trained the DeepSeek R1 AI system using 1,000 of its Ascend AI chips. The project builds on the open-source DeepSeek-R1 model, and avoids politically sensitive subjects in ordinary interactions, with the company claiming it to be ‘nearly 100% accurate.’ However, the success rate dropped to 40% in more complex tests, when users employ role-playing or disguise their intentions.

The project falls in line to regulate AI to comply with Chinese national values and established boundaries of expression. Previously, other domestic AI platforms, including Baidu’s Ernie Bot, already restricted responses to politically charged or sensitive topics. Huawei’s DeepSeek-R1-Safe also aims to formalise such controls within the LLMs’ advanced AI servers.

Reuters also reports that after its launch, the DeepSeek R1 model surprised its rivals with a cost-effective system and its open-source nature, as it offers unlimited free usage to most users globally. It is claimed that Huawei has spent $294,000 on training the R1 model, which is much lower than other popular AI chatbots from the West.