Morgan Lopes

June 24, 20261 min read

"Tokens" are the greatest Jedi mind trick of this decade.

"Tokens" are the greatest Jedi mind trick of this decade.
People talk about them like they're obvious.
Spoilers: they're not.

So what is a token?

They're not characters.
They're not words.
They're not sentences.

Sometimes, but not always, tokens are one character.
- "A"
- "?"
- "$"

Sometimes, but not always, tokens are a whole word.
- "the"
- "and"
- "because"

Sometimes it’s only part of a word.
- "under" + "standing"
- "play" + "ground"
- "inter" + "national"

... and sometimes a single emoji, a line break, or even a space can become its own token.

The exact definition also changes depending on which model you're using. Nope, they're not interchangeable.

Yet we've built an entire vocabulary around them.
- "How many tokens?"
- "What's your token limit?"
- "Can we reduce token usage?"

Weirder still, token count isn't a measure of complexity:
- Some incredibly difficult tasks require very few tokens.
- Some trivial ones require a lot.

But still, we squabble about them...
- We celebrate when we save a few thousand of them.
- We panic when we use too many.
- We price software with it.

My guess: Our future will criticize us for mistaking tokens for what mattered.

New posts by email

A short note when something new goes up. Never more than once a week.