LANDMARK · 6 MIN · USE
Ask, don’t invent
After this landmark, you can prompt a model to surface its uncertainty and ask for what it’s missing, rather than papering over gaps with confident invention.
Left to its defaults, a model would rather produce a smooth, complete-looking answer than admit it lacks a fact — so it fills the gap with something plausible. You can shift that behaviour by asking for it directly: “if you’re not sure, say so,” “ask me for anything you need,” “mark anything you’re inferring.” This doesn’t make the model truthful — it can’t know what it doesn’t know — but it visibly raises how often gaps come back as questions or flagged assumptions instead of silent fabrication. The instruction changes what a good answer is allowed to look like: incomplete-but-honest beats complete-but-invented.
This is the canonical concept. It stays the same across learner lenses so personalization never changes the underlying facts.
◌ EVERYDAY LENS
What this looks like for you
Add “tell me what you’d need to answer this well, or say if you can’t” to prompts about facts, dates, or numbers. A returned question is a save, not a failure.
MAKE A DECISION
You need a specific regulation’s effective date and add “if you’re not certain, say so.” The model gives a confident date anyway. What now?
CARRY THISAdd one clause to your next factual prompt: “ask me for anything you’re missing, and flag anything you’re inferring.”