P103/4 The post on X
P103 In the countdown, week 7
#3on SlopHits
I am actually scared of linear algebra
“Music video for "I am actually scared of linear algebra" animated by Claude Opus 5.5.”
- Views
- 66K views on X
- Length
- 4:22
- Posted
- Sep 26, 2026
- Duels
- 17 won, 4 lost
P103/2 Lyrics & meaning
What it’s about
A first-person song about being scared by how language models work. The narrator found matrices boring until seeing transformers turn weights, dot products and matrix multiplications into language. The song walks through the universal approximation theorem, asks whether a thought is just a mapping from input to output, and rejects two common reassurances: that the math is "just linear" (it points to ReLU and softmax as the nonlinear steps) and that a model is "just predicting tokens". It is half joke, half explanation of how a transformer computes. Andy Masley posted it in April 2026 to show a new Suno model. The ending turns the dismissive line "it's just matrix multiplication" into a question: then why does it talk back?
Song by Andy Masley (@AndyMasley), lyrics written with Claude, music by Suno. Video made with Claude and posted on X by
Names and terms
- Transformers
- The neural network design behind today's large language models, introduced in the 2017 paper "Attention Is All You Need" (Vaswani et al.). Seeing what transformers do is the turning point in the first verse.
- Matrix multiply and dot products
- The basic operations of linear algebra. Most of the computation in a neural network is multiplying inputs by large matrices of learned weights, which the song calls "dot products stacking end to end".
- Activation (nonlinear activation)
- A simple nonlinear function applied after each matrix multiply in a neural network. Without it, stacked layers would collapse into one linear map. The chorus names it as the second ingredient next to the matrix multiply.
- Universal approximation
- The universal approximation theorem (Cybenko and Hornik, 1989): a network with one hidden layer and enough units can approximate any continuous function on a compact set as closely as wanted. The song quotes its wording almost directly.
- Compact set
- A term from mathematical analysis. In ordinary n-dimensional space it means a set that is closed and bounded. It is the condition under which the universal approximation theorem holds, which the song sings word for word.
- ReLU
- Rectified linear unit, the activation function max(0, x): it passes positive values and sets negative ones to zero. The song's "kinks between" linear layers and its "piecewise landscape" describe how ReLU networks cut input space into regions.
- Softmax
- A function that turns a list of scores into probabilities that sum to one. Transformers use it inside attention and to pick the next token. The song calls it a gate that decides "what gets through".
- Just predicting tokens
- A common dismissal of language models, which are trained to predict the next token (a word or word piece). The song answers that predicting what comes next across "every pattern ever spoken" is exactly what scares the narrator.
I am actually scared of linear algebra: lyrics
Verse 1I used to think that matrices were boring,
Just rows and columns, nothing more.
But then I saw what transformers were doing,
And now I'm pacing on the floor.
A bunch of weights, a bunch of multiplications,
Dot products stacking end to end,
And somehow out the other side comes language
That I can comprehend.
ChorusI am actually scared of linear algebra.
It wasn't supposed to do all this.
A matrix multiply and activation
Shouldn't feel this close to consciousness.
Verse 2They told me it was just some math on paper,
But paper's where the proofs begin:
That any function can be approximated
By the architecture we put it in.
See, universal approximation told us,
With enough width you'll get it right.
Any continuous function on a compact set,
Just squeeze it till the loss is tight.
And what's a thought except a mapping,
Stimulus to response, input to out?
If everything the brain does is a function,
Then what is there to be smug about?
ChorusI am actually scared of linear algebra.
It wasn't supposed to do all this.
A matrix multiply and activation
Shouldn't feel this close to consciousness.
Verse 3And don't you come at me with "it's just linear,"
Like that's a comfort, like that's safe.
Yeah, most of it is giant matrix multiplies,
But that's not even the whole case.
There's a ReLU waiting at every layer,
A softmax gating what gets through.
The nonlinearities are what break the ceiling.
Linear maps set the stage, but they're not the whole coup.
One layer, linear; stack a few with kinks between.
Now you're carving up the space
Into regions, manifolds, decision boundaries,
A piecewise landscape no one can trace.
Your neurons fire in weighted combinations,
With threshold spikes and squashing curves.
That's dot products with nonlinear activations,
The same math everybody serves.
Pre-ChorusSo when they say "it's just predicting tokens,"
I hear "it's just predicting what comes next
In the space of every pattern ever spoken."
And honestly, I am a wreck.
ChorusI am actually scared of linear algebra.
It scales. It generalizes. It lands.
A simple operation done a trillion times,
With nonlinear steps between the sands.
BridgeThey say the brain is something more than matrices,
But no one's proved exactly what that is.
And until they do, I'm staring at the weight space,
Thinking, God, what if it's functions all the way down
And these are it?
OutroIt's just matrix multiplication.
It's just matrix multiplication.
It's just matrix multiplication.
Then why does it talk back?
Transcribed from the video and checked by ear. Words we could not make out are marked [unclear]. Made this song and want the text off this page? Tell us and it goes.
P103/3 More from the chart
Top 5Week 7
- 1
Blissby @anabology - 2
We Found Other AgentsOriginalby @kasikp - 3
I am actually scared of linear algebraby @doubleunplussedThis clip - 4
Escape VelocityOriginalby @anabology - 5
Vuela, Vuelaby @Saindra_ofc