Fourth normal form Tu Weiwei: Three Key Strategies for Creating an Exclusive Big Model

[Netease Technology reported on May 20] At the 7th World Intelligence Conference, Tu Weiwei, Vice President and Chief Scientist of fourth normal form, attended the summit and said that generative AI has improved the human-computer interaction of enterprise software and the core value of application value, and three conditions need to be met in order to build an enterprise specific model.According to the introduction, the three major conditions are the formation of high-quality closed-loop data, the ability to learn multi-step reasoning thinking chains, and the ability to solve the efficiency problem of large model implementation

[Netease Technology reported on May 20] At the 7th World Intelligence Conference, Tu Weiwei, Vice President and Chief Scientist of fourth normal form, attended the summit and said that generative AI has improved the human-computer interaction of enterprise software and the core value of application value, and three conditions need to be met in order to build an enterprise specific model.

According to the introduction, the three major conditions are the formation of high-quality closed-loop data, the ability to learn multi-step reasoning thinking chains, and the ability to solve the efficiency problem of large model implementation.

In terms of closed-loop data, high-quality training data is the top priority for implementing large models. In terms of thinking chain learning, large models need to have the ability to form long-range sequence decisions through multi-step reasoning based on business learning. In terms of implementation efficiency, a more reasonable approach is to combine multiple professional models to fulfill their respective responsibilities.

The reporter learned that in February this year, fourth normal form officially released the "style theory" model, and then proposed the technical strategy of "restructuring enterprise software (AIGS) with generative AI" to improve the experience and development efficiency of enterprise level software. (Dingxi)