Temperature
A setting that rescales next-word probabilities to control randomness — low values make output focused, high values more diverse.
When you interact with an artificial intelligence model, the temperature setting acts as a dial for creativity. It controls how random or predictable the AI's response will be by shifting the mathematical probabilities of the words it chooses next.
At a low temperature, between point-one and point-seven, the model plays it safe. It sharpens the probability distribution, making the most likely words even more dominant. The resulting text is focused, conservative, and highly repeatable. If you turn the temperature all the way down to near zero, the model always picks the single most probable word, leaving no room for surprise.
When you turn the temperature up, above point-eight, the model flattens those probabilities. This gives less common words a real chance to be selected. The output becomes more diverse, creative, and unexpected, though you run the risk of the text becoming incoherent. A temperature of exactly one leaves the model's learned probabilities completely unchanged.
For anyone tracking how a brand or concept is represented by AI, temperature is the key. It explains why the exact same prompt can yield a completely different description from one run to the next. By shaping token probabilities, the temperature setting ultimately decides whether the AI sticks to the script or wanders into more creative territory.
Temperature is a setting that controls how random or creative a model's output is by rescaling the probability distribution over the next word before one is chosen. It's applied as a divisor inside the softmax step: probability = softmax(logit / temperature).
Low temperature (say 0.1–0.7) sharpens the distribution, making likely words even likelier — output is focused, conservative and repeatable, and at values approaching zero it becomes greedy decoding that always picks the single most probable word. High temperature (0.8 and up) flattens the distribution so less likely words get a real chance, adding diversity and surprise at the risk of incoherence. A temperature of 1 leaves the model's learned probabilities unchanged.
For AI visibility, temperature explains why the same prompt can yield different brand descriptions on different runs. It works on the same token probabilities that top-p sampling filters, and both shape what our Tree Walker observes.