June 14, 2023

An API for Everything

Generally speaking, humans have created a word for the vast majority of things we care about.

Of course, a word isn’t the thing, it’s shorthand for the thing. It’s like a logo or a symbol or a map that represents the thing in a way that most people find useful in most cases.

For example, if you and I were chatting and I said “tree,” I’d have a picture in my head of a tree, and you’d have a picture in your head of a tree.

We’d almost definitely not both be picturing the exact same tree, but the word “tree” is a useful enough approximation in most contexts to enable us to continue with our conversation about landscaping the yard or building a tree house or whatever.

This “useful approximation” of words becomes incredibly interesting when you think about large language models (aka LLMs, which is what ChatGPT is).


Because, in a very practical sense, language is an API for most things that humans care about.

We certainly do not have a word for everything, but most of the things that most of us care about can be represented reasonably well with words.

We don‘t agree exactly on their meaning - like the tree example above - but in most cases, it’s close enough that it’s close enough.

Why does this matter?

This matters because if everything we care about can be represented with words, and computers can use LLMs to interact with words like an API, then LLMs provide an API for everything we care about.

This is incredibly powerful.

So powerful, in fact, that it’s hard for me to think of any aspect of society that won’t be transformed by it.