What generic AI gets wrong with this prompt
Columbia wrote the word dream into a goals prompt on purpose. Generic AI, trained on ten thousand cautious goal statements, edits the dream right back out.
It answers “dream job” with a job title
“My long-term goal is to become a general manager in consumer tech” — generic AI converts the dream into a LinkedIn headline, because safe goal statements are what its training rewards. But Columbia invited imagination. A dream job you can’t picture yourself inside — the problems on your desk, the people affected by your calls — reads as a destination you haven’t actually visited in your own head.
Ask your AI — “From my draft, describe one working day in my dream job. If you can’t fill a paragraph, where would that texture need to come from?”
It replays the resume the prompt just set aside
Columbia opens by saying it already has a clear sense of your path to date — the past is stipulated. Generic AI spends the first paragraphs re-proving it anyway, because your inputs are past-shaped and the model writes what it’s given. Every backward sentence here contradicts the prompt’s own instructions.
Ask your AI — “Highlight every sentence about my past. The prompt says Columbia already knows this — what forward-looking detail could each highlighted sentence become instead?”
It mistakes rank for a dream
Generic AI reads “dream job” as maximum title — CEO, partner, managing director — because models equate ambition with altitude. But a title is a chair. What makes a dream job a dream is the problem you’d finally get to own from that chair, and that problem is the one detail the average of everyone’s essays can never supply.
Ask your AI — “What problem does my dream job exist to solve, and for whom? If my draft only names the position, what is the position for?”