LANDMARK · 7 MIN · USE
Examples as guidance
After this landmark, you can steer a model with a worked example instead of abstract instructions — and recognise when an example misleads it.
Describing what you want in words leaves the model interpreting adjectives; showing it one or two finished examples pins down tone, structure, and level of detail in a way instructions rarely can. This is why a single “here’s the shape I mean” often outperforms a long list of rules. But examples teach by imitation, so the model copies whatever they carry — quirks, formatting, even a bias you didn’t intend. The skill is choosing examples that demonstrate the pattern you actually want, and noticing when the model has latched onto a surface feature instead of the point.
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
When a description isn’t landing, paste one example of a good answer and say “more like this.” Showing the shape beats stacking up more adjectives.
MAKE A DECISION
You give the model three example product descriptions — all for kitchen gadgets — and ask it to write one for a backpack. It comes back oddly kitchen-flavoured. Why?
CARRY THISNext time an instruction won’t land, replace it with one example of the output you want — and check what else that example might be teaching.