Red Teaming and Safety

DeepSeek reportedly does not do red teaming on models, contrasted with western AI safety focus, China's pragmatic approach to AI safety through usage regulation rather than model-level restrictions

← Back to Notes on DeepSeek

While some argue that Chinese AI regulation focuses on usage rather than the models themselves, users question whether this is a distinction without a difference in practice. Direct experiences with DeepSeek suggest that the model remains heavily restricted, as it consistently mirrors official political narratives or abruptly terminates conversations when sensitive topics arise. Ultimately, these observations suggest that regardless of the formal regulatory framework, the end result is a user experience defined by strict ideological guardrails and model-level censorship.

1 comment tagged with this topic

View on HN · Topics
I don't get the part of "AI models are not regulated directly, the government instead has restrictions on how those models can be used in software, services". Is it not the same thing? When I chat with DeepSeek about any (Chinese) political/social issue, it immediately begins aligning with the party's line or just cut off the conversation abruptly.