Baidu Innovates by Tailoring AI Models to Different Scenarios

On April 16th, at the Create 2024 Baidu AI Developer Conference, Baidu’s founder, chairman, and CEO, Robin Li, introduced a new approach. Baidu is capable of tailoring smaller models to various scenarios by considering factors such as effectiveness, speed, inference cost, among others. This customization supports fine-tuning and post-pretraining.

Robin Li mentioned that models created through this dimensionality reduction approach outperform directly extracted models from open source repositories of the same size. Similarly, for models with equivalent performance, the cost is significantly lower. He concluded, “This means open source models will gradually lag behind.”