By Eduardo Baptista, Yuvraj Malik
BEIJING (Reuters) – Baidu said its experience customizing its search engine to meet China’s regulatory requirements showed that its AI-powered chatbot could make mistakes on “important and sensitive topics.” The company said Tuesday it is confident it will not.
Baidu CEO Robin Li said in a call with analysts that the company is waiting for government approval before launching a ChatGPT-like Arniebot. A Reuters test found that the bot refused to answer a wide range of political questions, especially those about Chinese government leaders.
“For important and sensitive subjects, artificial intelligence should not hallucinate,” Lee said, using industry jargon for when AI models produce different output than expected.
“Given that LLMs (Large Language Models) are more or less probabilistic models, this task is perfectly fine,” he adds, noting that the models used by many AI chatbots such as ChatGPT and Ernie bot have Mentioned.
Lee said industry regulations are not yet final and the company will continue to update its strategy as it evolves.
“Baidu has operated search in China for over 20 years and has extensive experience with Chinese culture and regulatory environment,” he said. “Conversely, companies that do not have extensive experience in delivering relevant online content or a track record of working closely with regulators will face significant challenges.” Chinese cyberspace regulator last month released a draft measure to govern services driven by generative AI like Erniebot, and said the content generated by this frontier technology would align with China’s core socialist values. said it must be. Li said these measures would benefit Baidu. “We believe that early regulators’ active involvement in generative AI will raise the entry barrier, and we are well positioned to do so,” he said.
(Reporting by Eduardo Baptista, Beijing; Yuvraj Malik, Bangalore; Editing by Mark Potter)
Copyright 2023 Thomson Reuters.
