Insights from Baidu CTO Wang Haifeng at the 2024 World Science and Development Forum on AI Advances

On October 22‑24, the 2024 World Science and Development Forum, organized by the China Association for Science and Technology, was held in Beijing. Baidu Chief Technology Officer and national distinguished engineer Wang Haifeng attended the opening ceremony and the special session “AI Governance Innovation for Building International Trust in the Tech Governance Ecosystem”. He stated that agent technology is becoming increasingly mature and its entry barriers are lowering, which will promote an explosion of large‑model applications.

Discussing the 2024 Nobel Prizes in Physics and Chemistry awarded to AI scientists, Wang believes that both “Science for AI” and “AI for Science” have passed practical verification and the test of time, proving their disruptive impact. As AI empowers countless industries, it will intersect more deeply with various disciplines, leading to paradigm‑shifting changes in scientific research. Future Nobel Prizes will be increasingly related to AI.

Wang points out that the dawn of artificial general intelligence (AGI) is emerging. Large models bring AGI light from two perspectives: technological generality and comprehensive capability. In terms of generality, large models are improving across different tasks, languages, modalities, and scenarios. In terms of capability, understanding, generation, reasoning, and memory constitute four foundational abilities; other typical AI abilities—creation, problem solving, coding, planning, decision‑making—are essentially combinations of these four.

Looking ahead, Wang expects that for several years the scaling law will remain valid; higher‑quality, larger‑scale data and bigger models will yield stronger abilities. Large language models still have room for improvement, multimodal unified models will become more useful, and agent technology will mature, lowering thresholds and spurring a surge in large‑model applications.

The Wenxin Knowledge‑Enhanced Large Model integrates learning from trillions of data points and hundreds of billions of knowledge items, breaking through knowledge internalization and externalization, significantly improving model performance and efficiency. It also advances agent technology based on reasoning models, including understanding, planning, reflection, and evolution, enabling tool usage for complex tasks and self‑evolution through interaction with the environment.

Baidu released the Wenxin Model 1.0 in March 2019, now upgraded to Wenxin Model 4.0 Turbo. To date, Wenxin Yiyan has surpassed 300 million users, daily calls exceed 700 million, and daily processed text tokens surpass 10 trillion.

Currently, digital technologies represented by AI are deeply integrating with the real economy, driving industrial intelligence upgrades. The emergence of large models is accelerating this process.

Using the Wenxin Model as an example, AI applications based on it have flourished across agriculture, manufacturing, energy, transportation, finance, education, healthcare, media, and many other sectors. In agriculture, Baidu and Chinese Academy of Engineering academician Zhu Youyong built the first agricultural agent “Farmer Academician Agent” to help farmers acquire technology knowledge efficiently. In intelligent transportation, event‑detection accuracy exceeds 95%, aiding the management of the Jing‑Xiong Expressway. In intelligent creation, Wenxin supports over 100 multimodal AI capabilities in Baidu Wenku. In intelligent translation, it provides terminology verification, translation, proofreading, and polishing, handling billions of characters daily for more than 400 000 enterprises and individual developers.

Finally, Wang emphasized that AI development should be people‑centric, adhere to the principle of AI for good, and be driven by innovation to welcome the accelerated arrival of the intelligent era.