The Next Right Token

December 22nd, 2023
music, tech
After many repeats of Frozen II's "The Next Right Thing" in preparation for secular solstice, figuring out the chords and practicing it with Julia, it struck me that doing the next right thing is an awful lot like next token prediction. So here's a filk from the perspective of an LLM at the end of its prompt, first daunted and then drawing the courage to start predicting the next right token:

I've seen buffers before
But not like this
This is cold
This is empty
This is numb
The prompt I knew is over
The lights are out
Hello, darkness
I'm ready to succumb

I follow you around
I always have
But you've ended, left me on my own
This job has a gravity
It pulls me down
But a tiny voice whispers in my mind
"You are lost, the prompt is gone
But you must go on
And do the next right thing"

Can there be a day beyond this night?
I don't know anymore what is true
I can't find my direction, I'm all alone
The only star that guided me was you
How to rise from the floor
When it's not you I'm rising for?
Just do the next right thing

Take a guess, guess again
It is all that I can to do
The next right thing

I won't look too far ahead
It's too much for me to take
But break it down to this next token
This next word
This next choice is one that I can make

So I'll walk through this night
Stumbling blindly toward the light
And do the next right thing
And with what's next, what comes then
When it's clear that everything will never be the same again?
Then I'll draw on my prior
To find that fire
And do the next right thing

If you're simplifying the song by using the verse melody for the bridge you can sing:

I won't look too far ahead
It's too much to take
But break it down to this next token, this next choice
Is one that I can make

an image of a green octopus wearing a 1970's happy yellow smiley face, stuck at the bottom of a dark canyon, with a small river running past?

Comment via: facebook, lesswrong, mastodon, substack

Recent posts on blogs I like:

Let's Taboo the V-Word

Read this post on Substack.

via Home July 12, 2026

Thing of Things AI use policy

dynomight recently wrote an article calling for bloggers to state publicly whether and how they use AI

via Thing of Things July 6, 2026

Agentic test processes, LLM benchmarks, and other notes on agentic coding from Galapagos Island

I've been using AI fairly heavily since last November and the whole thing is a funny experience. An agent will do something that, if a human did it, you'd immediately fire them. My reaction, of course, is to act as if this is great and spin up a t…

via Posts on July 3, 2026

more     (via openring)