“The core ideas behind modern Large Language Models were substantially invented between 2018 and 2019, but it took until late 2022 for a system to be built, tested, and proven useful.”