During his speech at the Create 2024 Baidu AI Developer Conference on April 16th, Robin Li, the founder, chairman, and CEO of Baidu, pointed out that in the future, natural language will become the new universal programming language. "You just need to know how to speak, and you can become a developer," he said. "AI is sparking a creativity revolution, where developing applications in the future will be as simple as making a short video. Everyone will be a developer."
At the event, the official release of WENXIN's Large Model 4.0 Tool Edition was showcased. Li Yanhong announced that since the launch of WENXIN Yiyan over a year ago, the user base has surpassed 200 million. "Compared to a year ago, the algorithm training efficiency of WENXIN's large model has increased by a factor of 5.1, with a weekly training effectiveness reaching 98.8%, and the inference performance has improved by 105 times, reducing the cost of inference to only 1% of the original," he stated.
During the conference, Li Yanhong shared Baidu's specific approach to developing AI-native applications based on large models, highlighting three noteworthy directions: Mixture of Experts (MoE), small models, and intelligent agents. He remarked, "This is the result of countless trials and errors and the payment of high tuition fees over the past year."
Li Yanhong firstly emphasized that in the future, most large-scale AI-native applications will be based on MoE. He explained, "Here, MoE refers not to the general academic concept, but to the hybrid use of large and small models, not relying on a single model to solve all problems."
Secondly, he discussed small models. Small models have low inference costs and fast response speeds. In specific scenarios, finely tuned small models can achieve comparable performance to large models.
Thirdly, he mentioned intelligent agents. Li Yanhong noted, "Intelligent agents" are currently a hot topic, and with the improvement of their capabilities, they will continuously give rise to a plethora of AI-native applications. The mechanism of intelligent agents includes understanding, planning, reflection, and evolution. It enables machines to think and act like humans, autonomously complete complex tasks, continuously learn in the environment, and achieve self-iteration and self-evolution.
Looking ahead, Li Yanhong believes that multimodal large models, including the fusion of text, images, speech, and videos, are a crucial long-term development direction for foundational models. They are the necessary path towards Artificial General Intelligence (AGI).
