Chinese AI chatbots reportedly must answer 2,000 questions, prove censorship compliance
For American companies building AI today, it’s basically a free-for-all, a self-regulation zone with zero federal restrictions.
But for Chinese AI companies, the Chinese Communist Party exerts strict control over what models get released and what questions they cannot answer.
A report in The Wall Street Journal details the rigorous tests that AI models are subjected to before being released on the global stage to compete with Western AI models.
AI models must answer 2,000 questions that are frequently updated and achieve a 95% refusal rate for queries related to forbidden topics, like the Tiananmen Square massacre or human rights violations, according to the report.
The strict regulatory framework does have some safety advantages, such as preventing chatbots from sharing violent or pornographic material as well as protections from self-harm, an issue that American AI companies are currently wrestling with.
A report in The Wall Street Journal details the rigorous tests that AI models are subjected to before being released on the global stage to compete with Western AI models.
AI models must answer 2,000 questions that are frequently updated and achieve a 95% refusal rate for queries related to forbidden topics, like the Tiananmen Square massacre or human rights violations, according to the report.
The strict regulatory framework does have some safety advantages, such as preventing chatbots from sharing violent or pornographic material as well as protections from self-harm, an issue that American AI companies are currently wrestling with.
