Baidu's AI Summit Sheds Light on the Opportunities and Challenges of Large Language Models

On August 16, 2023, the WAVE SUMMIT Deep Learning Developer Conference 2023 took place in Beijing, China, hosted by the National Engineering Research Center for Deep Learning Technology and Application. During the conference, Wang Haifeng, Chief Technology Officer of Baidu, shared that Baidu's self-developed deep learning framework, named "PaddlePaddle," has gained significant traction, attracting 8 million developers and serving 220,000 enterprises and institutions. Moreover, Baidu has created 800,000 models based on PaddlePaddle.

One notable achievement is the development of ERNIE Bot (Chinese name: Wenxin Yiyan), a large language model by Baidu. ERNIE Bot has introduced five plugins on Baidu Search, Document, Image and Video. Those plugins enable the model to generate real-time and accurate information, provide long text summaries and question answering capabilities, offer data insights and chart production, facilitate image-based creation and question answering, and even perform text-to-video conversions. The plugin mechanism expands the boundaries of large model capabilities and caters to various scenarios.

PaddlePaddle and the ERNIE series of large models are two pivotal elements in Baidu's pursuit of building an artificial intelligence ecosystem.

Wang Haifeng believes that large language models have now acquired fundamental artificial intelligence abilities such as understanding, generation, logic, and memory. This signifies the advent of general artificial intelligence (AGI) and marks the transition of artificial intelligence into the stage of industrial mass production.

During the past six months, the AI large model industry has experienced super rapid growth. Major technology companies have not only made their foray into this field but have also engaged in fierce competition concerning product implementation and ecosystem development. They are providing integrated services for large models to third parties, complete with a comprehensive development toolchain.

According to Baidu's insights, amidst intense market competition, it has identified the "enterprise office" as a crucial entry point. For example, the adoption of large model technology has led to fundamental changes in enterprise work systems. Many functions can now be accessed through large model plugins. Taking the coding scenario as an illustration, Baidu Smart Cloud has developed a next-generation coding assistant tool called "Code Assistant Comate", built on the ERNIE large model.

During the software development process, Comate can combine code with contextual information and comments. It predicts code by analyzing declared function names and provides suggestions for viewing, as well as manual editing of the suggested code. It automates the completion of repetitive code segments and simplifies the workflow for engineers.

It is worth noting that Baidu has been utilizing various artificial intelligence platforms internally since this April.

On the evening of August 31 local time, Baidu announced that ERNIE Bot app had attracted over 1 million daily active users on its first day of availability for download.