Based on their revolutionary capabilities in information comprehension and text generation, generative large language models (LLMs) have been widely applied in production and daily life. The deep integration of their technological applications with social structures not only carries ideological attributes but also poses the risk of multi-dimensional proliferation regarding political direction, value orientation, and public opinion guidance. Driven by the practical need to address the ideological risks of generative LLMs, it is urgent to uncover their ideological attributes—from macro-theoretical interpretation to micro-application observation—and specifically analyze the causes of these risks. Furthermore, effective approaches to countering these ideological risks should be explored across three dimensions: ideological awareness, institutional safeguards, and technological measures.
Luo Shengyang (Thu,) studied this question.