Baidu’s ChatGPT Rival Ernie Bot Didn’t Go Live at Its Launch Party

The only thing missing was a live demo of Ernie Bot itself.

The presentation drew sharp criticism from online viewers who were disappointed not to see Ernie Bot performing live. While
Microsoft Corp.
and
Alphabet Inc.’s
Google have also relied on prerecorded demos, many commentators contrasted the event to OpenAI’s real-time demo two days earlier of its latest model, GPT-4.

Investors also balked: Baidu’s Hong Kong-traded stock price plunged as much as 9% after the presentation started, before closing 6.4% lower.

“Everyone will be benchmarking Ernie Bot today against ChatGPT, even against GPT-4—that bar is really high,” Mr. Li said at the event. “You could say that of all the world’s biggest companies, Baidu was first to release.”

Microsoft used OpenAI’s technology, while Google,
Meta Platforms Inc.
and
Amazon.com Inc.
have yet to release similar products, he said. Ernie Bot will be available to invited users from Thursday, Baidu said.

Mr. Li acknowledged at the beginning of his presentation that many people had asked him why the company rushed to launch the technology and wondered whether it was truly ready.

“During our initial internal testing, it’s true I wouldn’t describe the experience as perfect,” Mr. Li said. But the market demanded it, he said, adding that “everyone is waiting for this technology.” 

Ernie Bot represents a high-stakes gamble for Baidu, which has pressed employees since the start of the year to develop the product as quickly as possible. The chatbot could solidify Baidu’s reputation as an AI leader, a label it has repeatedly assigned itself in recent years to regain investor favor after tumbling from the elite tier of China’s tech giants. 

In pushing something out quickly, the company sought both to capitalize on the wild popularity of ChatGPT and beat out its domestic competitors.

The initial market reaction, however, hearkened instead back to Google’s debut of its AI chatbot Bard, which it also rushed to push out in the face of OpenAI and Microsoft’s challenge to its core search business. In one day, the American search giant lost $100 billion from its market capitalization after Bard produced factual errors in a demo.

During Baidu’s event, Mr. Li showed videos of Ernie Bot’s capabilities, including its fluency with Chinese idioms and ability to generate audio speaking in different Chinese dialects. Mr. Li said the bot would be great in five areas—literary creation, business writing, mathematical calculation, Chinese-language understanding and multimodal generation, such as generating images or videos from texts.

“Ernie Bot isn’t a tool for China-U.S. technology confrontation,” he said, “but a product of generations of Baidu technicians chasing the dream of changing the world with technology, a brand new platform for us to serve hundreds of millions of users and empower thousands of industries.”

Online, some people said Baidu wouldn’t have shied away from chatting with Ernie Bot live if it had mature capabilities.

“I get it, it’s not ready yet, just a pitch deck!!” wrote one viewer on the event’s English-language live stream on YouTube. “Two words: that’s it,” wrote another on the Chinese Twitter-like platform Weibo, adding: “What was on display is just a patchwork version of Baidu Baike,” referring to the company’s Wikipedia-like product.

Some Baidu employees who weren’t involved in the final stages of Ernie Bot’s development said they, too, were skeptical of the product’s capabilities and would need to test it before passing judgment on its quality.

One business attendee at the event was more optimistic about the product’s business utility. He said his firm could consider using Ernie Bot, such as by having it read through various documents to generate summaries as well as give titles to videos that his company produces or handles. Such usage could improve worker productivity, the attendee said, adding that he expects Ernie Bot to keep improving.

Once seen as one of China’s shiniest tech giants, alongside
Alibaba Group Holding Ltd.
and
Tencent Holdings Ltd.
, Baidu’s standing took a hit after its core advertising revenue began to slow and the company reported a loss in 2019.

It has since struggled to regain investor confidence. Baidu’s market cap has largely remained below $100 billion since 2014 while Tencent’s and Alibaba’s grew to above $800 billion in early 2021.

Casting around for new sources of nonadvertising revenue, Baidu turned to its existing AI talent and technologies. In 2019, it elevated

Wang Haifeng,
a natural-language processing expert who was then the director of its AI and technology groups, to become its first chief technology officer in nine years. It pumped money into a wide range of technologies, including chip and software infrastructure for AI development, self-driving cars and cloud computing.

In 2019, it developed a deep-learning model known as Ernie, repurposing for Chinese language a breakthrough algorithm that Google had introduced two years earlier. Baidu subsequently developed dozens more Ernie models, including for image and art generation, similar to those of OpenAI’s Dall-E.

Baidu struggled to commercialize those technologies, but the sudden release and virality of ChatGPT late last year changed the equation. It opened up a window of opportunity for Baidu to finally cash in on its years of investment. ChatGPT shares the same DNA as Ernie’s AI models: Both are so-called large language models built on the same algorithmic foundation that came out of Google.

In late December, as buzz grew over ChatGPT, Mr. Li emphasized to employees the importance of commercializing the new advancement. By early January, company executives had mobilized employees to start pulling together a ChatGPT-like product with its Ernie models, The Wall Street Journal reported.

After months of pent-up anticipation, Mr. Li didn’t specify on stage when the chatbot would become widely available. Baidu initially said last month that it would embed the chatbot into its search engine and open it to the public in March.

Some employees told the Journal that the tight timeline has been insufficient for developing a well-functioning product.

Up until the final days before the launch, the team was still rushing to ready the chatbot, leaving no time left for a planned internal rollout to all employees to perform companywide testing, according to people familiar with the matter.

Write to Karen Hao at [email protected], Yoko Kubota at [email protected] and Raffaele Huang at [email protected]