Why large language models are mysterious – even to their creators

Created to accompany an exhibition at the Computer History Museum in California, this nifty explainer from the video essayist Grant Sanderson (aka 3Blue1Brown) helps to demystify how large language models (LLMs) operate – even though, as Sanderson details, the inner workings can be mysterious even to those who program them. Offering a brief history of their evolution up to today, he deconstructs both the hidden scaffolding of these technologies and the innovations that have allowed them to improve so dramatically over the past decade. With a narrow focus that avoids controversies around the ethics of LLMs, Sanderson focuses on what’s actually happening when one interacts with these technologies. Through this lens, he elucidates how LLMs are an intricate – and rather ingenious – collaboration between human and machine inputs.