Zhipu Founder on GLM-5.3: Scaling Laws Not Just Parameter Scale, Post-Training Becomes Key Variable

Zhipu AI founder Tang Jie stated in an article that large model scaling should not focus solely on parameter count, but also needs to consider data scale, compute resource allocation, inference costs, and actual operating conditions. The article reviewed the shift from Kaplan Scaling Laws to Chinchilla Compute-Optimal Theory, pointing out that the early path of "parameter growth faster than data growth" led to resource misallocation in some super-large models, while subsequent research indicates that model parameters and training data should grow in a more balanced manner.