The Impact of Large Models' Success on Data Modeling

The principle of parsimony in data modeling is facing challenges with success of large models. To address this challenge, the computing theory of large models is studied from the perspective of data computing in the paper. It has been found that the fast algorithms for small-sized and medium-sized models may not necessarily hold true, but in the large models, the fast algorithms for small-sized and medium-sized models become valid. Then, it is pointed that large models have the properties of “easy to compute” through case analysis. Finally, a principle shifts in data modeling are proposed, Data modeling must consider the computability principle, followed by the principle of parsimony. Therefore, the success of large models is scientific inevitability, not the result of brute force.