Stephen Downes

Take your time and give this paper a nice slow read. As the author says, "No single idea is beyond the abilities of a smart teenager to understand. But what is beautiful and surprising and remarkable is that the phenomena we observe in LLMs is not magic but simply the emergence of a complex system from simple rules." Yes, it seems hard to imagine that these language models can really work. But clearly they do, and this paper describes the step-by-step process over the years as researchers discovered how the could. Via Data Science Weekly 660.