I remember a time, and it wasn’t that long ago, when telling someone in an interview that you were good at training models or working with generative AI could get you the side eye, followed by a very polite no-callback.
Just after Stable Diffusion landed, having diffusion or segmentation work on your resume could turn a normal technical interview into an ethics tribunal with coffee.
You’d get grilled on data governance, model provenance, consent, training sets, bias, overfitting, copyright, deployment risk, and every other concern in the drawer.
Some of those questions were fair, others were theater.
Especially when the interview's supposed to be about their CRM.
But the mood was real: “AI person” was not always a safe professional identity for a technical generalist like me.
I’m glad those days are mostly over.
Computer vision especially, is some of the most creatively satisfying work you can do. It sits at the intersection of math, perception, systems design, and the quirks and cues that quietly tell you what’s wrong with what you just did. At least, if you know how to listen to it.
Agents are fun to work on for similar reasons.