Play it like this — few-shot prompting.
You can describe the sound you want. Or you can just play a couple of examples and let the model follow.
Last lesson you spelled out the job in words. There's a sharper way to get the phrasing you're after: play the model a few examples of the job already done, then ask. Asking cold, with no examples, is zero-shot; laying down a handful first is few-shot.
A phrase is a short, ready-made run of notes a player keeps handy and drops into a solo. Picture handing the player two you already like — "over C7, play it like this; over F7, like this" — both written the same way. Faced with a fresh chord, they answer in the same shape.
It works because the model is, underneath, a pattern-continuer — the idea from Lesson 1. Give it a pattern and it carries the pattern on; nothing is learned or stored, the examples just tilt what comes next. And a worked example fixes the shape, the note choices, even the feel far more exactly than a sentence of description can.
Show the player zero, then one, then two example phrases — and hear the new answer snap into the same bluesy shape.
A couple of examples, then the ask.
In code, few-shot is just a few worked pairs placed before your real request. The model reads the pattern and carries it on.
solo = model.play( system = brief, examples = [ ("over C7", "C Eb E G"), # root, b3, 3, 5 — bluesy ("over F7", "F Ab A C"), # same shape, moved ], ask = "over G7", ) # → "G Bb B D" — the same shape, continued
Show, don't just tell.
Describing the phrasing you want is fine; playing it is better. A couple of worked examples settle the questions a sentence leaves open — the shape, the notes, the feel — and the model just continues the pattern it sees.
Reach for few-shot when the model nearly does what you want but not in the right shape. Show it two or three clear phrases — the tricky moves included — and let it copy.