It's saves space in the context window. ChatGPT has a large template that question go into that is (largely) what enforces the morality stuff. You still want t

It's saves space in the context window. ChatGPT has a large template that question go into that is (largely) what enforces the morality stuff. You still want the model to have the training on morality since users will want responses based on different systems of ethics.