For most of last year, I was in conversations about AI without really understanding what I was talking about. I would say 'large language model' or 'transformer' or 'context window' and the room would nod and the conversation would move on, and I would tell myself I would go look it up later.
I never went to look it up. I kept nodding.
It bothered me. Not because I was getting away with it, I was. Because I was advising people on how to use a tool I could not have explained the basics of. There is a kind of credibility you build by sounding fluent. There is a different kind you build by actually being fluent. I had the first one and I was running on it.
Late last year I started forcing myself to learn the basics. Not the math. The shape. What is a token. What is pretraining. Why does the model not 'know' the way a person knows. Why does it confidently make things up. What does the temperature dial actually do to the output.
Sounding fluent buys you the meeting. Being fluent buys you the next ten.
I was surprised how much of it was reachable for someone with no CS background, if I went slowly and stopped pretending the bits I did not understand were unimportant. The hard part was not the material. It was the willingness to ask basic questions in a room of people who had already moved past them.
What I have noticed since: the rooms I am in feel different. Not because I now know everything. I do not. Because I am no longer paying the small mental tax of pretending.
If you are reading this and you are also nodding along, I am not coming after you. I was you. The exit is the same exit I took. Pick the smallest piece you do not understand. Stop pretending. Go learn it.
What were you nodding along to last year?