(Yicai) July 4 -- Chinese developers of large language models should not compete in technology but in applications as use brings technology alive, according to Robin Li, chairman of Baidu.

Open-source and closed-source models are both useless without application, the chief of the Beijing-based search engine giant said at the World Artificial Intelligence Conference in Shanghai today. “In the era of AI, applications that are 'super capable' are more significant than 'super applications' which only appreciate daily active users.”

Benchmark models are gradually entering workflows. For example, it takes LLMs 19 seconds to deal with express delivery orders instead of the usual three minutes, Li said. Moreover, programmers at Baidu are getting used to using a co-pilot as 44 percent of such code works and by now around 30 percent of all Baidu code is generated by AI.

Baidu released its own LLM Ernie Bot in March 2023. The ChatGPT-like model has 500 million daily users, up from 200 million logged two months ago, reflecting real demand, per Li.

The industry veteran touched on the topic of sharing source code. Closed-source models are more powerful than open-source ones with similar parameters and scale because those with shared source code require more parameters to catch up, which results in higher reasoning costs and slower response, he explained.

Li still admitted that open-source models are valuable in some specific scenarios such as academic research and teaching. However, as the AI race intensifies, companies need to compete in efficiency through savings, he said, adding that in this case, commercial closed-source models are superior.

The tech billionaire said that he is most optimistic about the development of AI agents as various types of such autonomous helpers are expected to appear in fields such as healthcare, finance, education, manufacturing, transportation, and agriculture, generating an ecosystem of over one million agents.

Editor: Emmi Laine