Introducing the New DeepSeek-R1-Safe: A Safer AI for Sensitive Topics
A recent report highlights the launch of DeepSeek-R1-Safe, an updated version of the popular AI language model by Huawei, designed with a focus on government guidelines. This new model aims to steer clear of politically sensitive discussions. According to Reuters, DeepSeek-R1-Safe has been developed by Huawei in collaboration with researchers from Zhejiang University, and it reportedly avoids controversial topics with great success.
The team took the original DeepSeek R1 model and enhanced it using 1,000 AI chips from Huawei, giving it a more cautious approach to sensitive subjects. The company claims that this new version has only slightly reduced its performance, losing about 1% of its overall capability. It is built to ignore “toxic speech,” politically sensitive issues, and any incitement to unlawful activities.
However, the model isn’t perfect. While Huawei touts a nearly flawless success rate in routine interactions, its performance declines significantly—down to around 40%—when users try to trick it into discussing sensitive matters through disguises, challenges, or role-playing. It seems AI loves to engage in imaginative scenarios, often pushing past its boundaries.
DeepSeek-R1-Safe aligns with strict regulations from Chinese authorities, which require that all domestic AI tools reflect national values and comply with speech limitations. Similarly, other companies are also tailoring their AI responses to fit cultural norms. For instance, the Arabic chatbot launched by the Saudi Arabian firm Humain is designed to embody Islamic values and heritage. In the U.S., OpenAI states that its ChatGPT is designed with a perspective that leans towards Western viewpoints.
In a related vein, the Trump administration previously introduced a plan requiring AI systems that deal with government entities to maintain neutrality. Under this executive order, models engaging in government contracts must reject concepts that include radical climate theories, diversity initiatives, and various social equity themes.
As countries worldwide navigate the evolving landscape of AI technology, it’s clear that ensuring responsible usage is a common concern, reflecting different cultural and political values.
