On April 16, at the Baidu AI Developer Conference, Robin Li, founder, chairman and CEO of Baidu, revealed the progress of Wen Xin Yi Yan one year after its release. The number of users has exceeded 200 million, and the average daily API calls have exceeded 200 million. , the number of customers or enterprises served is 85,000. After one year of development of the Wenxin large model, the inference cost has dropped to 1% of the original level. 27% of Baidu’s new codes every day are automatically generated by Comate, an intelligent code assistant based on the Wenxin model.
Robin Li also said that in the future, large-scale AI native applications will basically be MoE (mixed expert model), which means there will be mixed applications of different sizes and must be adapted according to specific scenarios. In some specific scenarios, the efficiency of a small model after fine adjustment can achieve the effect of a large model. Not long ago, Robin Li said in an internal speech that whether it is China or the United States, the strongest basic models are closed source, and various small models and the best small models are distilled from large models. The model produced by reducing the dimensionality of a large model is better.