That’s not even true. LLMs in their modern iteration are significantly enabled by transformers, something that was only proposed in 2017.

The conceptual foundations of LLMs stretch back to the 50s, but neither the physical hardware nor the software architecture were there until more recently.