The new flattening hides in whoever writes the prompts
Hook
When AI summarises a thousand comments, someone decided how to sort them. Who is that, and can anyone challenge the rules?
Claim
If one small team decides how a system clusters opinions, what counts as a topic and which documents are linked, it has become the new cadastre. The grid has moved from the form to the prompt, where it is harder to see. Rules of interpretation must be open, versioned and arguable like any public rule.
Story
Petar’s essay names this the risk that worries him most, along with confidently wrong answers (a wrong answer with a footnote is more dangerous because the footnote looks like proof) and counting only what is easy: votes per councillor are easy, whether a neighbourhood trusts its council is not. No named person yet; the story is the choice an engineer makes when setting how comments are grouped.
Sources
- Petar’s own essay
Check before posting
- A concrete, real example of a prompt-driven distortion would carry this. None found yet.
- Needs a person-level story.
Counterargument
Any human summary has the same problem. The gain is that these rules can be inspected.
Post angles
Facebook (Croatian)
- Kad umjetna inteligencija sažme tisuću komentara, netko je odlučio kako ih složiti. Tko je to, i može li mu itko osporiti pravila?
- Odgovor s fusnotom opasniji je od odgovora bez nje, jer fusnota izgleda kao dokaz. Zato izvor mora biti na jedan klik, a netko ga mora i kliknuti.
LinkedIn (English)
- When AI summarises a thousand comments, someone decided how to group them. Who is it, and can anyone challenge the rules?
- A wrong answer with a footnote is more dangerous than a wrong answer without one, because the footnote looks like proof. The source has to be one click away, and somebody has to click it.