Baidu to Launch Chinese ChatGPT-Style App and Conversational Search
The Google of China is not waiting after the OpenAI hit from 2022
Baidu is the search giant of China. Like Google, it generates the majority of its revenue from its position as the leading search engine in the country. Unlike Google, it is trying to get out in front of upstart rivals that might introduce a ChatGPT-like service.
Bloomberg reported this week, citing an anonymous inside source, that Baidu plans to introduce a ChatGPT-style app and incorporate large language model (LLM) technology into search in March 2023.
Chinese internet search major Baidu Inc is planning to launch an artificial intelligence (AI) chatbot service similar to OpenAI's ChatGPT in March, a person familiar with the matter told Reuters.
The technology firm plans to launch the service as a standalone application and gradually merge it into its search engine, said the person, who declined to be identified as the information is confidential.
A Generative Giant in China
Baidu is no novice when it comes to LLMs and generative AI. The company claimed at its annual developer conference in December 2022 that it had released more than 20 large-scale AI models. The models range from text and image generation to specialty applications for specific “industries such as energy, finance and biocomputing,” according to a company press release.
This is all built on top Baidu’s open source AI development platform PaddlePaddle, which was introduced in 2016. The company’s LLM, and GPT-3 equivalent, is called ERNIE. It was incorporated into PaddlePaddle in 2019 and will likely be the technology behind the new applications and features.
The new app may also speak via Baidu’s virtual assistant Xiao Du or through its virtual human character Xijiajia. Baidu announced new tools to create virtual human clones of real people for metaverse virtual worlds at an event in July 2022. It intends to enter an era where “every human being will be able to possess a virtual counterpart.” It would make sense for Baidu to fuse these two technologies and introduce an interactive conversational virtual human based on ERNIE.
Conversational Search
It is also notable that Baidu may add LLM features to search. Many LLMs, including GPT-3, are not well suited to search. They do not keep track of the sources of their training data. However, there are some options to combine generative LLMs with retrieval models similar to what Perplexity.ai has done to create that link back to the sources most responsible for the generated text.
This also enables the new benefits of conversational search. A true conversational search experience maintains context throughout a session and can answer questions similar to what you might expect from a human expert. The interaction design also enables users to refine their search to get better results if the AI model misinterprets an initial intent or if they want more detail around specific information in a search result.
Microsoft is expected to add this type of feature to its Bing search engine after integrating GPT-3 into the service later this year. In that case, Bing may serve as the retrieval model, and the GPT-3.5 or GPT-4 model may focus on interpreting the request and generating the text response after receiving data back from the search engine.
A Parallel AI Universe
China is a parallel digital universe in many regards. The Chinese government operates its own internet, voice assistants in China tend to be from local tech giants and not those from western countries, and everything from popular consumer devices to telecom equipment tends to originate in the country.
What is somewhat different in the AI market is that so much of the research, and many of the foundational AI models, are in the public domain. Other technology segments tend to be driven by more proprietary solutions. This is not to say AI doesn’t have its share of proprietary technology. However, innovation can come from anywhere, and the next big AI advance may come from China. Some of that may make it directly into other solutions outside of the country or inspire new innovation.
Regardless of how this plays out, you should expect a full complement of Chinese solutions that match up closely to solutions rolled out in the U.S., Europe, Japan, and South Korea. Some of them are bound to launch first in China. While it is popular to talk about Chinese companies simply copying innovations from the western world, that is not likely to be the case with AI. Chinese companies and research labs have been working on the very same problems at the same time.
While ERNIE has some similarities to GPT-3 and also Google LaMDA, and Deepmind’s Sparrow, it is not a copy of these solutions. Granted, many companies both in China and the rest of the world may copy the ChatGPT interface because that has been such a hit with users.
The ChatGPT Effect
Many articles have highlighted the challenge that ChatGPT has introduced for Google. LaMDA has many of the same features as ChatGPT and is even available in a limited preview through Google’s Test Kitchen app. However, it is not widely accessible, and the interface and scope of use cases are more limited today.
Google appeared to be planning a more deliberate rollout of many of its AI products. That changed because ChatGPT captured the popular imagination, and suddenly Google became viewed as a laggard, even if that perception is untrue from a technical perspective. The company is now trying to accelerate all of its programs to recapture some mindshare and begin to secure some market share before OpenAI gains too much momentum.
Baidu has some insulation with its position in the Chinese market but is not immune from these types of disruption. The move to introduce a new app and features to search is an attempt to be proactive before a homegrown LLM emerges and threatens its dominance in search and other digital services.
