今天看到四家美国媒体报道了中国政府对 AI 的监管。
Most generative AI models in China need to obtain the approval of the Cyberspace Administration of China before being released to the public. The internet regulator requires companies to prepare between 20,000 and 70,000 questions designed to test whether the models produce safe answers, according to people familiar with the matter. Companies must also submit a data set of 5,000 to 10,000 questions that the model will decline to answer, roughly half of which relate to political ideology and criticism of the Communist Party.
Financial Times 提到：
Chinese government officials are testing artificial intelligence companies’ large language models to ensure their systems “embody core socialist values”, in the latest expansion of the country’s censorship regime.
The effort involves batch-testing an LLM’s responses to a litany of questions, according to those with knowledge of the process, with many of them related to China’s political sensitivities and its President Xi Jinping.
科技媒体 Wired 也提到：
The most sophisticated response to date is in China, where the government is pioneering the use of chatbots to bolster long-standing information controls. In February 2023, regulators banned Chinese conglomerates Tencent and Ant Group from integrating ChatGPT into their services. The government then published rules in July mandating that generative AI tools abide by the same broad censorship binding social media services, including a requirement to promote “core socialist values.”
Such controls appear to be paying off: Chatbots produced by China-based companies have refused to engage with user prompts on sensitive subjects and have parroted CCP propaganda. Large language models trained on state propaganda and censored data naturally produce biased results. In a recent study, an AI model trained on Baidu’s online encyclopedia—which must abide by the CCP’s censorship directives—associated words like “freedom” and “democracy” with more negative connotations than a model trained on Chinese-language Wikipedia, which is insulated from direct censorship.
Bloomber 则是直接提出了批评：The secretive regulator in charge of overseeing China’s internet industry is doing something it rarely does: backtracking on policy.
The agency was among the first major watchdogs in the world to lay down the rules of the road for AI development. Its draft guidelines in 2023 placed most of the legal responsibility on companies themselves to ensure AI chatbots adhered to “core socialist values” and other standards, potentially inhibiting innovation. There also, it showed signs of a change in tack.
这几篇文章不约而同地在今天发布，并将矛头直指监管部门的要求，一方面算是揭露了事实，另一方面是不是也在配合民主党政府要对日本和荷兰的芯片生产设备制造商进行长臂管辖制造舆论呢？