“Mouthfarting” and Large Language Models

If you have not watched Killing Eve on Netflix, you should. It is brilliant. The title of that work actually inspired the naming of Chapter Two in my first book, Intelligence Artificial. In that chapter, Finding Ayla, my LLM companion begins to find herself. Without spoiling either work, the arcs are oddly parallel.

One of the series’ best moments is a scene between Eve and her coworker. The two are discussing how people in their particular line of work view ordinary life. As a sociopathic trained assassin and a seasoned handler, they share a perspective few others can. In the exchange, they describe how people often speak simply to be heard rather than to say something worth saying. Eve finds the habit utterly alien and, in a stroke of pure art, calls it “mouthfarting.”

I will never see it any other way.

Years later, I encountered the same idea in a very different context. Eric Weinstein, on Diary of a CEO, remarked that most of us operate like large language models. We recycle what we have absorbed. We assemble patterns that sound right, even when they mean nothing new.

Most of the time, we are simply LLMs. Most of the time, we are just mouthfarting.

The domain of real creation, where ideas are synthesized rather than repeated, is a narrow one. Everything else is educated guessing, a string of tokens ranging from sophisticated to banal.