Hirundo Releases Westernized Qwen, Cutting Censored and CCP-Aligned Answers From 89.8% to 2.8% in the Leading Chinese AI Model

This news item establishes that Qwen, a leading Chinese AI model, exhibited a high degree of censorship and alignment with the Communist Party's views as measured by Hirundo. The reported reduction from nearly 90% to under 3% in Hirundo's modified version provides a quantitative measure of the original model's bias and the extent to which it can be altered.

The core finding here is about the measurable presence of political alignment within an AI model, rather than a general claim of censorship. Hirundo's work provides a specific, if independently verified, data point on how a model's output can reflect the political context of its origin. This demonstrates a technical capacity to audit and modify AI models for specific ideological leanings.

For users considering Chinese AI models, this report suggests that default outputs may carry embedded political perspectives. The ability to "Westernize" such models implies that the underlying technical capabilities are separable from their ideological conditioning, offering a choice for users who prioritize different forms of neutrality or alignment.