Han Guan/AP
Baidu’s shares fell as much as 10% on Thursday after the first release of China’s response to ChatGPT released only a pre-recorded video of AI chatbot Ernie, a web search firm.
The Beijing-based tech company claims Arnie will reshape its business, and for weeks has been discussing plans to incorporate generative artificial intelligence into search engines and other products.
But the millions of people who attended the event on Thursday had little idea whether Baidu’s chatbot could compete with ChatGPT.
At Ernie’s highly publicized and highly anticipated press conference, Baidu founder Robin Li stood next to an open chat screen and recounted questions previously entered into the chatbot. He admitted that he had only shown a demo of the technology the company had prepared earlier.
Li said some users will be able to test Ernie themselves soon, but didn’t provide a timeline for the full public rollout. The company is starting with a limited opening of the business to his partners.
Arnie’s planned launch comes as US groups such as OpenAI and Google continue to push the development of generative AI forward. OpenAI released its latest AI model, GPT-4, this week. It claims it can beat some humans on tough professional exams such as the U.S. bar exam.
Microsoft is incorporating GPT-4 into Bing and other products, and US companies are testing the technology in their workplaces and products.
Nearly every Chinese tech giant is vying to roll out domestic versions of ChatGPT, and analysts believe Baidu’s long-standing investments in AI and development of natural language learning technology have given the group an early lead.
“Sometimes I use it and am pleasantly surprised, and other times I think there is an obvious error,” Li told the audience. “But one thing is for sure, it’s going very fast.”
But live comments posted by many of the two million people watching Li’s speech on the WeChat social media app showed disappointment, as information about the fall in Baidu’s stock dominated their feeds. It didn’t take long to get there.
“A very strong ability to understand and express language allows any company to get closer to their customers,” Li said. “This is an opportunity for all businesses, and it can even affect everyone.”
At the end of the approximately 45-minute demonstration, a chorus of WeChat users raised their voices in various ways. Two of his Baidu employees said they were equally disappointed by the development.
For months, the company has been diverting internal resources to building Ernie, including most of Baidu’s servers running Nvidia’s powerful A100 chips, according to two people familiar with the AI team. It also included reassigning to
OpenAI’s release of ChatGPT last November caught Baidu off guard, two people close to the company’s AI efforts said, adding that by that point the US group had significantly superior technology. He said he didn’t believe he had.
According to people close to Baidu, the launch will see Baidu catch up with and implement the painstaking process of fine-tuning Arnie with human feedback.
This training program requires humans to evaluate thousands of responses given by an AI model. This allows the AI model to gradually increase its score and improve. One of his employees said that due to the relative lack of public disclosure about how OpenAI conducted its training, Baidu’s team had to conduct its own experiments.
The US group also said it would not release details about the GPT-4’s technical underpinnings, as it is a change from previous models such as the GPT-3, and has released some information.
“We can only explore by ourselves. It took OpenAI more than a year to train ChatGPT, and another year to tune GPT-4,” said a Baidu employee. “That means we are two years behind.”
Baidu did not immediately respond to a request for comment.
Additional reporting by Nian Liu in Beijing.
© 2023 The Financial Times Ltd. All rights reserved. Do not redistribute, copy or modify in any way.
