AI Basics: How to Understand Your Power Drill and Stop Being Afraid of It

The first and most basic thing to understand about LLMs is that they are statistical models of language. That’s it.

Some people see their incredibly impressive output and say, Statistics? It’s more than just statistics!

But it really is just statistics.

And that’s why you can start using it and stop hiding under your bed.

LLMs Learn Language Similarly to Humans

When humans are born, they know no specific words. They acquire language simply by listening to how others use it.

AIs learn language in just the same way; they are given no information about how language works, just the words themselves.

This reveals one profound truth:

1. The ability to command English is largely encoded in how English is used.

And we make two conclusions:

1. These LLMs can develop an astonishingly sophisticated command of language—absorbing, manipulating, and regurgitating vast amounts of human knowledge.
2. This is the only kind of learning AI does.

LLMs Learn Language Differently Than Humans

Human language learning involves more than just associating words with other words.

When a child learns the word red, they aren’t just learning how it relates to other color-words. They already know what red is—they’ve seen it.

When children learn the words “one” and “two”, they don’t just learn how to use number-words in context. They already know the true meaning of “one” and “two”—they experienced that meaning five minutes ago when there were two toddlers and one toy.

Concepts like love, hate, justice, friendship, anger exist independently of the words we use to describe them.

What AI Math Fails Teach Us

LLMs are famously bad at arithmetic.

Given how AI learns, this suggests that learning “math” is very different than learning “language”.

In other words, the language-learning strategy has limitations.

These limitations are easy to see with math, where the output from your AI is provably and obviously wrong. But we need to draw the deeper lesson: These limitations are actually pervasive. And if you’ve used AI extensively, for actual work, you’ve encountered these limitations often enough.

AI Is More Impressive Than You Think and Less Impressive Than You Think

This is why it is (currently) shortsighted to be in terror of the coming AI apocalypse.

Whenever true understanding is required—where meaning isn’t just about how words relate but about something deeper—LLMs will fall short.

The more you use AI, the less you’ll be afraid of it.

This might sound like the soothing, naive voice of a blind pragmatist, but it could also be the voice of a friend, telling you that it really is okay to retire your horse and buggy and buy a power drill.