I don’t need the source code to randomly change bytes in the compiled Linux kernel binary either.

Fine-tuning weights is easier than retraining the foundation model from scratch with a different corpus.

Source code makes it easier to modify a binary, training data doesn't make it easier to modify a trained network.