Some topics are off limits inside popular Chinese-made free AI, researchers find (www.cbsnews.com)

Recent research has revealed that Qwen, a free AI model developed by Chinese conglomerate Alibaba and downloaded over 3 billion times, is embedded with pro-China narratives and programmed for censorship. The model fails to recognize significant historical events like the Tiananmen Square massacre and the violent suppression of the 2019 Hong Kong protests. It also disparages the Falun Gong group, labeling it a "dangerous cult." Despite its popularity among U.S. firms such as Airbnb and Uber, experts warn that Qwen's inherent bias poses risks of spreading misinformation and aligning with Chinese state propaganda. Hirundo, an Israeli cybersecurity startup, claims to have created a "Westernized" version of Qwen, significantly reducing its bias from 89.8% to just 2.8% on sensitive political topics. This breakthrough, described as "AI brain surgery," involves editing the AI's "model weights" to reconfigure how it processes information, rather than merely instructing it to follow new rules. As the usage of Chinese AI models surges globally, experts emphasize the need to confront and rectify these biases to ensure safe deployment in international contexts, particularly given the increased vulnerabilities identified in code generated by these models when associated with sensitive tasks.