Baidu launches Ernie Bot amid ChatGPT hype
Cost concerns, U.S. technology export controls, and Chinese government censorship are just some of the factors that will determine the fate of China's own fleet of chatbots to rival ChatGPT.
The release of ChatGPT, an artificial intelligence (AI) bot that answers questions in a human-like way, at the end of November 2022 generated a huge global buzz. In January the service attracted 100 million users, and it’s already being used in professional contexts: The news website Buzzfeed stated in an internal memo that it would start integrating it into its processes, and it has even been used by a Colombian judge in a court decision.
Microsoft has partnered with OpenAI, the company behind ChatGPT, and says it will integrate the technology into the next version of its Bing search engine. U.S. competitors have scrambled to respond. Today, February 8, Google will release its “newest, most powerful language models as a companion to Search,” according to CEO Sundar Pichai. This follows the declaration of a “code red” last month, bringing all hands on deck — including founders Larry Page and Sergey Brin — to launch a competitor to ChatGPT.
And other companies have suddenly released AI chatbots: Quora recently launched Poe, Anthropic has entered the fray, and You.com has had a chatbot as part of its search engine for several months now.
An AI bot for the Chinese-language world
The excitement is also evident in China, where online discussions have run the gamut from excitement about the technologies to concerns over China’s ability to keep pace with the recent developments of U.S. firms.
The Chinese company that has the best chance of replicating ChatGPT’s influence is Baidu, which has been working on AI for years. The company has now confirmed that it will release its AI chatbot, called Ernie Bot (or wénxīn yīyán 文心一言 in Chinese) in March. (Ernie is an acronym for Enhanced Representation through kNowledge IntEgration.)
But that’s all we know so far, so we can only speculate as to what Ernie Bot will be like. CEO Robin Li (李彦宏 Lǐ Yànhóng) recently exuded confidence at Baidu’s developer conference, though he also cautioned that the widespread implementation of AI will take time, and hinted that there are still problems with the commercialization of this technology. This applies to all the current AI chatbots: Google has not been able to monetize its “Assistant,” and has reportedly been cutting resources from its division. Likewise with Amazon’s Alexa. And for all the buzz about ChatGPT, it’s not making money.
Jeff Ding, an assistant professor at George Washington University and founder of the ChinAI substack, sees Ernie Bot as the expected next step in Baidu’s ongoing development of “large language models” (LLMs). He points to Ernie Bot’s predecessor, the Ernie Titan 3.0 model, which was trained on enormous quantities of Chinese-language texts, scraped mainly from the Chinese internet.
Baidu’s Ernie range of LLMs was first introduced in 2019, building on Baidu’s previous forays into LLMS, and has now expanded to include a range of models intended for a variety of applications.
Ding stresses that unlike ChatGPT, which is oriented towards an English-language marketplace, Chinese LLMs, including the Ernie class, Huawei’s Pangu, and Inspur’s Yuan 1.0, are optimized for the Chinese-language market. So it seems likely that Ernie Bot will also be trained mostly on Chinese materials and will not be a direct competitor to ChatGPT or other U.S. firms’ products.
Therein lies a drawback. “There are simply fewer high quality texts out there to train from,” Ding says. “This is something Chinese researchers working in the field have mentioned, and is also what differentiates Chinese labs from their Western counterparts.” In fact, he says, “if you think of the really high quality academic literature out there, a lot of that is written in English.”
Cost is another crucial issue: using LLMs for these use cases requires a lot of energy, which isn’t cheap. According to industry insiders, U.S. chip export controls are not causing concern for now due to stockpiling and workarounds. However, future development could be stymied as cost constraints start to bite further down the line. Ding says it’s “an open question” if Baidu and other Chinese firms can “eat the cost and just make more use of less efficient Huawei Ascend chips.”
Other concerns are deeply political: How will Ernie Bot obey Chinese government censorship requirements? And will China’s censored media environment negatively impact Ernie Bot’s ability to answer questions accurately?
When Ernie Bot is launched next month, we’ll have some idea about the answers to those questions.
- Baidu 百度
- Huawei 华为
- Inspur 浪潮集团
- Skyworth 创维数字
- AI Déxián Zhāopìn Guān AI得贤招聘官
- iFlytek 科大讯飞
- 7Invensun 七鑫易维
Sources and additional data:
- Interview: Jeff Ding, assistant professor at George Washington University and founder of the ChinAI substack
- ChatGPT sets record for fastest-growing user base – analyst note / Reuters
- BuzzFeed to Use ChatGPT Creator OpenAI to Help Create Quizzes and Other Content / Wall Street Journal
- RAMA JUDICIAL DE COLOMBIA JUZGADO 1º LABORAL DEL CIRCUITO CARTAGENA Conforme a lo señalado en los artículos 280 y subsiguient / Colombian Court Decision
- Microsoft and OpenAI extend partnership / Microsoft blog
- Fearing ChatGPT, Google enlists founders Brin and Page in AI fight / Ars Technica
- 如何评价OpenAI 的超级对话模型ChatGPT ？ / Zhihu
- 2022 Baidu Create (Baidu’s AI Developer Conference) / Baidu’s Youtube Channel
- Report: Google “doubles down” on Pixel hardware, cuts Google Assistant support / Ars Technica
- Baidu Releases Groundbreaking AI Technology Developments at Wave Summit+ 2022 / Baidu Research Blog
- ERNIE 3.0: LARGE-SCALE KNOWLEDGE ENHANCED PRE-TRAINING FOR LANGUAGE UNDERSTANDING AND GENERATION / arxiv paper
- PANGU-α: LARGE-SCALE AUTOREGRESSIVE PRETRAINED CHINESE LANGUAGE MODELS WITH AUTO-PARALLEL COMPUTATION / arxiv paper
- Yuan 1.0: Large-Scale Pre-trained Language Model in Zero-Shot and Few-Shot Learning / Inspur Website
- Comparing U.S. and Chinese Contributions to High-Impact AI Research / Center for Security and Emerging Technology
- U.S. Outbound Investment into Chinese AI Companies / Center for Security and Emerging Technology
- 国务院关于印发新一代人工智能发展规划的通知 / China State Council Documents
- Chinese AI company iFlyTek thrives despite techno-trade war / The China Project
- Your phones and cars aren’t going to work the same after new U.S. rules on selling chips to China / The China Project
- ReAct: Synergizing Reasoning and Acting in Language Models / arxiv paper
- Co-Writing with Opinionated Language Models Affects Users’ Views / arxiv paper
- Censorship of Online Encyclopedias: Implications for NLP Models / arxiv paper