The Improviser · Lesson 5 — The Daring Dial ← Course

The daring dial — temperature.

Ask the improviser to finish the same phrase twice and you may hear two different notes. One dial — temperature — sets how often that happens.

Back in Lesson 1, the player never really faces a single certain note. It faces the odds: a spread of candidate notes, each with a probability. Yet it has to commit to exactly one. How it chooses is the question this lesson answers.

It chooses by drawing from those odds, and a setting called temperature governs the draw. Turn temperature down and the player almost always takes the single likeliest note, so the line comes out steady and repeatable. Turn it up and the rarer notes get a real chance, so the phrasing starts to vary and surprise.

Picture a cautious soloist who ends every phrase the same safe way — the home note, every single time. A daring one, facing that identical moment, might reach for something more unexpected. Temperature is the dial between those two players.

Turn it low when you want consistency — a fixed answer, a set format, a dependable ending. Turn it high when you want range — fresh ideas and variety. It is the same temperature knob you would set on a real chatbot, and it changes nothing the model knows: same player, same phrase, only how boldly it commits.

Set the temperature, then have the player draw a short line over the same C-major phrase. Low and safe, or high and daring — watch the odds reshape, and hear it.

Set a temperature, then play a line drawn from these odds.
safe daring 0.7
BALANCED
odds over the next note · at this temperature
a line of 8 notes drawn from these odds
no line yet — press play

One setting on the draw.

Temperature is a single argument. Low for a line you want the same every time; high when you want the player to range around the rarer notes.


      

The dial between steady and daring.

The player turns the phrase into odds, then draws one note from them. Temperature sets how that draw behaves: low, and it keeps taking the top note — the safe favourite, the same resolution each time; high, and the rarer, spicier notes start landing — varied, fresher, less predictable.

It's still the odds from Lesson 1 — temperature only sharpens or flattens them before the draw. It is not the player's confidence: the underlying scores are identical at every setting. Turn it down for consistency, up for range — but remember that more variety is not the same as more right.

Quick check · your ear
You set the temperature very low and ask the player to resolve the same phrase eight times. What do you hear?
Next: the other dials →
Go deeper — what temperature really does optional

It reshapes the odds

Under the hood, temperature divides the raw scores before they become probabilities — the softmax step from Lesson 1's maths. A low temperature stretches the gaps apart, so the top note towers over the rest; a high temperature squeezes them together, so the notes even out and the draw gets a real choice. The odds you met in The Odds were always tunable — this is the dial.

Greedy at zero, wild at the top

Pushed all the way toward zero, temperature stops being a draw at all — the player simply takes the single most likely note every time (this is called greedy, or argmax, decoding), which is fully predictable. Push it above 1 and the odds flatten toward an almost even pick. Everyday playing sits in between, so lines feel natural rather than robotic.

Variety is not quality

A higher temperature is not "more musical taste" — it is only a willingness to pick lower-probability notes. That can find a fresher phrasing, but it can just as easily reach for a note that clashes. Cranked too high, the wrong notes land — the model's version of a hallucination, playing something that doesn't fit with conviction. If a line must be correct and consistent, keep the temperature low.

Not the only dial

Temperature is not the only dial on the draw — the next page has two more.