Baidu Is Developing Its Own ChatGPT, Aiming to Integrate It Into Search

Beijing-based Baidu is joining a global race to commercialize a technology that has generated buzz among both consumers and businesses. The company, whose growth isn’t keeping up with that of its Chinese internet peers, has been pushing to refashion itself into an AI company, investing billions of dollars in technologies including self-driving cars and chips designed to power AI applications.

Chief executive

Robin Li

touched on ChatGPT in a late-December speech to some employees, saying it represents new opportunities, according to a transcript on Baidu’s internal website that was seen by The Wall Street Journal.

“We have such cool technology, but can we turn it into a product that everyone needs?” Mr. Li said, referring to AI-driven technologies including the chatbot. “This is actually the hardest step, but also the greatest and most influential.”

Bloomberg earlier reported on Baidu’s plans.

Trained on vast amounts of text data from the internet, ChatGPT is capable of answering all manner of user questions in fluent conversational prose. But the chatbot can’t guarantee accurate answers and at times has delivered sexist or racist comments, industry researchers have said.

Tech giants including

Microsoft Corp.

and

Alphabet Inc.’s

Google are hurrying to mature the technology underlying ChatGPT and integrate it into their products—including search, where its conversational abilities are seen to have the most potential to upend the status quo.

Microsoft, which invested in OpenAI in 2019 and 2021, announced fresh backing—as much as $10 billion, The Wall Street Journal reported—to infuse ChatGPT into its product lines, including its search engine Bing.

A similar Google technology known as LaMDA made its debut in 2021, and

Meta Platforms Inc.

released a chatbot known as Blenderbot in 2020, but neither has revealed plans to integrate the technology into their platforms.

Baidu has focused on an area of AI research known as natural-language processing, which can be used to develop voice assistants and chatbots like ChatGPT, using its vast repository of text data from its search-engine business.

In 2019, Baidu developed a system known as Ernie—a deep-learning model similar to the technology underlying ChatGPT—that it has used to make its search results more relevant. It has since developed dozens more Ernie models and extended their capabilities to include image and art generation, similar to those of OpenAI’s Dall-E. 

Baidu is now using Ernie as the foundation for its chatbot, and is training it on both Chinese- and English-language sources inside and outside China’s firewall, some of the people said. In the past Baidu has trained Ernie using sources that include Wikipedia, BookCorpus, Reddit and Baidu’s ecosystem of products—such as Baidu Baike and Baidu News—according to its open-source research papers.

Search results in China are censored under the state’s censorship rules.

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