But LLMs are not unchangable, there are several mechanisms available by now to drive change and learning (especially as context windows improve)
Practically for outcomes it doesn’t matter if that learning is happening in the LLM or on a layer above it.