To put even more simply, and IMO more interestingly:

LLMs are compression algorithms