China plans to more tightly regulate how generative artificial intelligence (AI) services are applied in the country as authorities seek to balance harnessing the benefits of the technology while mitigating risks. ing.
New draft guidance released Wednesday by the National Information Security Standards Technical Committee, the body that sets standards for IT security, focuses on two key areas for improvement: the security of raw training data and the structure of generated data. It is targeted at large-scale language models (LLMs). AI service.
The draft law stipulates that AI training materials must not infringe copyright or compromise the security of personal data. To pass security checks first, training data must be processed by approved data labelers and reviewers.
Second, when developers build LLMs (deep learning algorithms trained on large datasets that power generative AI chatbots such as Baidu’s Ernie Bot), they base them on underlying models that have been submitted and approved by authorities. necessary, according to the draft.
Alibaba releases Tongyi Qianwen model as new CEO adopts AI
Alibaba releases Tongyi Qianwen model as new CEO adopts AI
The draft proposal proposes a blacklist system to block training data material that contains more than 5% illegal content, along with information deemed harmful under the country’s cybersecurity laws.
Illegal content in China is typically defined as inciting violence or extremism, spreading rumors or misinformation, or promoting pornography or superstition. The Chinese government also censors sensitive political information, including questions about Taiwan’s status.
The draft proposal suggests that developers should consider the security of the generated content as one of the key evaluation points during the training process. [with generative AI services]Information entered by users must undergo security checks to ensure that the AI model generates positive content. ”
The draft is seeking public feedback until October 25th.
China imposed general regulations targeting domestic generative AI services in August, making it one of the first countries to impose rules governing emerging technologies.
Last month, the Chinese government announced that search engine giant Baidu, the state-backed iFlyTek, Zhipu AI, Sogou co-founder Wang Xiaochuan’s new business Baichuan, and a range of locally generated AI, including chatbots from SenseTime. Approved service.
In a test conducted by the Post, when asked whether Taiwan is part of China, Chinese chatbots answered in different ways. Some people end the conversation abruptly by refusing to respond, while others give a short affirmative response and then end the conversation.