“It seemed like a miracle: You put in your problem and through a process you didn’t have to think about, the right answer came out. I was floored by the apparent power of the thought process in this wide, low, blue-and-gray box—its flashing panel an artificial furrowed brow. I wondered how to apply the brain power in the box to my personal problems.” — Richard Gabriel

This was written in the early 1970s, when using an IBM 1130 for the first time. LLMs are the first thing that appeared in my lifetime (1985-present) that generate this sort of reaction.

What is stunning is that this “machine” is a machine that uses existing machines for your purposes. And that its interface is chiefly text (or voice, so I should just say “natural language”), on the screen of one of the ubiquitous digital hardware devices we’ve had around for decades.

(Interesting side note: the IBM 1130 was (I just learned) the cheapest IBM machine in its day. It had strong influence on pioneers like Guy Steele and Chuck Moore. Another example of the low-end being the avant-garde.)

(Tangential thought: I wonder if a “worse-is-better” case could be made about natural language as an interface. Low-bandwidth, linear, compared to interfaces that make more intense use of the visual cortex. And somehow, language wins again.)