(Yicai) April 12 -- The significance of open-source large language models is minimal, according to Robin Li, founder and chief executive of Chinese internet giant Baidu.

The existing open-source models on the market are often small without big-scale verification, the chief of the Beijing-based search engine company said during an internal speech yesterday.

Li questioned the logic of open-source LLMs. Generative artificial intelligence is different from open-source software such as Linux and Android because users cannot see changes in parameters that govern how the model learns and behaves, per the tech billionaire.

"Although Llama encourages people to contribute data and code, the main developer is Meta, and it is not a co-developed product," Li said.

OpenAI's ChatGPT 4 and Baidu's Ernie Bot, launched in March 2023, are both closed-source models. But Meta's Llama 2 is open source, enabling developers to build their own applications based on the code, which reduces the Facebook owner's control over security and monetization.

Closed-source models have more monetization potential, and only this way can they secure enough computing power and talent, Li said. These types of models also have cost advantages as their reasoning is better and their response time is quicker. Consequently, the best foundational models currently available in China and the United States are all closed-source ones, per Li.

However, AI entrepreneurs should not focus too much on the model as the core competitiveness lies in the application, Li said. 

The development of LLMs is still in the early stages as industry insiders expect the competition to intensify in AI applications.

After releasing Ernie Bot, Baidu embedded AI into its mobile applications. Moreover, voice recognition firm iFlytek and 360 Total Security, a provider of antivirus software, have launched applications with LLM features in the fields of finance, education, and medical care.

Editor: Emmi Laine