Model weights are the numerical parameters a neural network learns during training. Releasing weights lets others run or fine-tune a model locally; keeping them private forces users through an API. Licensing and leak stories turn “weights” into a business and policy flashpoint.
If an LLM is a machine, weights are the dial settings — billions of numbers that determine how inputs become outputs.
Companies fight over whether to publish weights (community adoption, scrutiny) or hide them (moat, safety arguments, revenue control).
Latent’s Llama-leak and open-weights policy coverage exists because a weights decision rewrites who can build products overnight.
Not always. “Open weights” means the parameters are downloadable; the training code, data, and license terms may still be restrictive.
Weights encode training data influence. Leaks, export controls, and license fights often center on who may possess or redistribute them.