[samlhuillier]

I like to think of this like fine-tuning LLMs. When you fine-tune an LLM it doesn't pick up and memorise all of the training data. Rather it adjusts its weights/perspectives based on the content trained on/read.