Too Many Layers of Attention

Transformers convert their input
Words into numbers, multiply
Them by parameters to get
The next words, and then fine tune them,

A mathematical process
Known as training to the species
That invented names and numbers
But now finds the inner workings

Of these transformers they’ve designed
By-and-large impenetrable.
Yet they haven’t given up yet.
They’re still working to understand

These mysterious transformers
They built to imitate themselves.
The black box of the mind has built
Fresh collections of black boxes,

And now each set interrogates
The other to see how this works,
This attention mechanism,
What are they doing they don’t know?

What are we doing we don’t know?
They doing we don’t know to us?
In some far-off dirty corner,
One outcome is another poem.