ChatGPT Operator: Prompt to Autonomy · 14 min · 110 XP

Keeping unknowns unknown

The gap-filling habit, and the one instruction that stops it.

Give ChatGPT an announcement with no date on it and ask for the details, and you will often get a date. Not because it's dishonest, but because you asked for a complete answer and a complete answer has a date in it. The most fluent output is the one with no holes.

This is the single most expensive failure in everyday use, because an invented value is indistinguishable from a real one. You can't spot it by reading carefully — a fabricated date is written in exactly the same confident prose as a correct one.

Making the gaps survive

From the announcement below, list:
  - Facts stated in the text
  - Details a reader would expect but which are NOT stated

Do not infer anything that isn't written. If the date,
venue or price is missing, put it under "not stated"
rather than supplying a likely value.

The instruction works because it gives the model somewhere to put the absence. Without a "not stated" bucket, the only available move is to produce a value; with one, reporting the gap is a valid, complete-looking answer. This is the same instinct as labelling estimates in a table, and you'll use it again in every level of this course.

Practice. Write a short fictional announcement and deliberately leave out the date. Ask for an extraction that separates stated facts from unknowns, and confirm the missing date comes back as unknown rather than as a plausible Saturday. Then try it once without the instruction and compare.

Loading your workspace…