Clicked Gallery

What is Temperature in AI?

Highlighted from a real engineering doc. Explained by Clicked.

Used in a sentence

Engineering Notes · AI Systems

Setting the temperature to 0.2 makes completions more deterministic, which suits extraction and classification tasks.

The reader highlighted one word in the docs. Clicked explained the technical term “temperature” in plain language:

Explained in three depths

Same facts, different vibe — Slang mode 😎

The Clicked way

●○○

Overview

Temperature is a setting that controls how predictable an AI model's word choices are. Low values make it pick the most likely next word almost every time; higher values let it gamble on less likely ones. It tunes variety, not intelligence.
●○○

Overview

Temperature is the chaos dial. Low means the AI plays the safest next word every time; high means it's willing to gamble — same brain, different appetite for risk. 😎

A quick take — often all you need.

●●○

Detail

Before writing each token, the model holds a ranked list of candidates with probabilities attached, and temperature reshapes that list. Near 0, the top choice wins almost every time, so the same prompt gives nearly the same answer twice. Around 1, the model samples in proportion to the original probabilities, and above that the gaps shrink until longshot words get real chances. Typical products run between 0.7 and 1.0 for chat, while code tools and extraction pipelines sit near 0, because a creative answer to a formatting task is just a wrong answer. Higher settings raise variety and with it the odds of drifting into nonsense, since every unlikely word makes the next one likelier to be unlikely too. The name comes from physics, where heat is what shakes particles out of their most stable positions.
●●○

Detail

Under the hood the model ranks every possible next chunk with odds attached, and temperature squeezes or flattens those odds. At 0 you basically get the same answer every run, which is why code tools and data extractors live there: creativity in a spreadsheet field is a bug. Chat products cruise around 0.7 to 1.0, enough spice to not sound fridge-manual flat. Crank past that and longshot words start winning, and each weird pick makes the next pick weirder, which is how you end up with confident essays about nothing. So no, temperature isn't a smartness slider, it's a dial for how often the AI picks the long shot. 😎

Want more? One click digs deeper.

●●●

Analogy

An office lunch order. Play it safe and everyone just gets the restaurant's bestseller, identical and reliable every single day. Turn the dial up and the intern picks by throwing darts at the menu: some days a discovery, some days deep-fried regret.
Deepest level reached.
●●●

Analogy

Spotify at a party. Locked to the greatest-hits playlist, every party sounds the same and nobody complains. Let autoplay take over and it starts guessing: sometimes a hidden gem, sometimes whale sounds clearing the room by song three.
Deepest level reached.

Unfamiliar concept? A real-world example makes it click — fresh analogies on tap.

AI explanations may contain errors · Not professional advice

Formal definition — The same term, explained the usual way

Temperature is a scalar applied to a model's output logits prior to the softmax, scaling the sharpness of the resulting probability distribution over candidate tokens. Values below 1 concentrate probability mass on high-likelihood tokens, increasing determinism; values above 1 flatten the distribution, increasing sample diversity at the cost of coherence.

Want Clicked to explain terms like “temperature” directly in your browser — including on PDFs?

Add to Chrome — Free

50 free Explanations · No credit card required