I got sort of pushed into working with AI for some work related tasks a while back (referring to generative AI in the form of chatbots, as appears to now be the default sociological meaning of the term). At first, I was somewhat intrigued – and curious, as usual – to talk to a computer, but also soon a bit underwhelmed with the results of my interactions. Little did I know that math of all things was not the strong suit of LLMs. 🤦🏼

However, after a dip in enthusiasm I soon began to realize that the systems are quite broadly knowledgeable. In hindsight, I see generative AI as the (my) third stage of knowledge access.

When I was in school, I realized that with the library I could get my hands on basically any book ever written. Maybe not literally, but almost. I remember that you had to pay a small postage fee to order any book that was ever published in Germany from the state library (which was located in Frankfurt, if memory serves.) I was not really a big reader, but I liked the idea and I was curious. Some of my nice interests in my late teens, such as Zen-Buddhism, were only accessible this way. I didn’t live in a big city.

Anyway, this access to information was fantastical to me. And that was level 1.

While I was at college, or music conservatory in my case, we had PCs with 56k modems to dial in and go online. It was exciting, and unimaginably slow. My English got better because I wanted to read books published in English and because I listened to a ton of lectures, seminars, and all kinds of material which I “discovered” on the internet. That was level 2.

Today, we have “a sort lossy zip file of the internet” as Andrej Karpathy called it. (I’m going to ignore the ethical problems of how we got there for this post.) Now we are able to navigate latent space. It sometimes feels like working with a hologram of human knowledge. The process is more exploratory than reliable. But I’m not going to lie: as frustrating that part can be, it is also rather intriguing for a curious-minded person. That is level 3.