Module 2 · Hook
But why does the same model swing from slop to sharp? To steer it, you have to know what it actually is.
It's not thinking. It's predicting.
The model doesn't know things. It predicts the next piece, one at a time, from what you gave it. That single fact explains everything: it's why context works, and it's why the model can be confidently, fluently wrong.
The job
When the model states wrong things with total confidence, I want to see how it actually produces words, so that I can catch and prevent hallucinations before they cost me.