[AINews] Death of Params: Z.ai CEO Jie Tang on GLM 5.3 and the new Post-training Scaling Law
Source
latent.space
Date
Context
Z.ai's account of the GLM-5.3 release illustrates why model size alone cannot describe a capability change. The company says it reused GLM-5.2's base model and improved it through further reinforcement learning on more varied, extended tasks. Its primary release article describes training environments in which a model must investigate a problem, make changes and verify the result.
The company also describes agents generating runnable tasks and checking that they are solvable, with synthesized reward signals for a subset. The supported point is that training environments and feedback can change performance without increasing the base model's size; the account does not establish a universal scaling law.
Terms in this piece · Glossary
inference — Running a trained model to get answers — the phase where AI is actually used, as opposed to trained.
