The weights, calculated through very intensive computing, are what hold the knowledge in LLMs, the source code just executes those. These products could just update/patch their weights periodically, and no one would complain because that's not bad per se.