Baidu, the Chinese search engine and artificial intelligence giant that recently made its chatbot accessible to the general public, has announced its upcoming launch of the latest iteration of its large language model (LLM). This move signifies a renewed show of support for China's thriving artificial intelligence (AI) industry, especially following the approval of the initial wave of ChatGPT-style services by Beijing.
Robin Li Yanhong, the founder and CEO of Baidu, revealed the upcoming introduction of the Ernie 4 model during a conference organized by the company's cloud computing division in Beijing on Tuesday. He lauded it as "among the top-tier Large Language Models (LLMs)."
According to Li, the latest version of Baidu's AI chatbot, Ernie 3.5, has processing speed twice that of the previous version with 50% improved efficiency.
Baidu's goal is to build a strong foundational capability for large models, support the development of native AI applications, and "hope that entrepreneurs can create blockbuster applications in the AI era on top of Ernie 4 model," said Li Yanhong.
Baidu announced last week that it will launch its generative artificial intelligence (AI)-powered chatbot, Ernie Bot, to the general public starting August 31, making it the first Chinese company to do so.
Referring to the China Science Daily publication, Baidu announced that Ernie 3.5 outperformed ChatGPT in several benchmark assessments.
In Chinese language evaluations, Ernie 3.5 also outperformed GPT-4, OpenAI's latest and more advanced model, as reported by the scientific journal.
The Chinese science journal referenced a test centered around standard admission and qualification exams, akin to college entrance or legal certification assessments. In this context, Ernie 3.5 demonstrated superior performance compared to both ChatGPT and GPT-4 when it came to the Chinese language.
Baidu Cloud also unveiled an upgrade to its model-as-a-service platform, Qianfan, and introduced a new ecosystem partners program.
Enterprises looking to develop large model applications may be more interested in the Qianfan Large Model Platform
Compared to the public-facing Wenxin Yiyan, enterprises looking to develop large model applications may be more interested in the Qianfan Large Model Platform. Qianfan is an all-in-one enterprise-level large model production platform launched by Baidu Intelligent Cloud in March of this year. It includes large model services based on Ernie or third-party open-source large models. The toolchain and development environment of the Qianfan Large Model platform will assist third-party companies in developing their own exclusive large models.
Qianfan now houses 42 Large Language Models (LLMs) sourced from both domestic and international providers. These include Meta Platform's Llama 2 and Google's Flan, among others, in addition to models developed by the Beijing Academy of Artificial Intelligence, according to Shen Dou, President of Baidu Cloud.
Baidu claimed that over 150,000 clients have already registered to utilize Qianfan, and nearly 10,000 of them are actively employing the platform to train their custom AI models and create applications for approximately 400 different business scenarios. These scenarios span a wide range of sectors, including finance, digital government, heavy industries, marketing, education, and transportation.