Temperature
Temperature is a setting on a language model that controls how random its word choices are: low values make it pick the most likely next word every time, high values let less likely, more varied options win.
No jargon, no math
In plain words, what is it?
Every time a language model writes a word, it holds a ranked list of candidate words with scores (logits). Temperature reshapes those scores before the dice are rolled. Divide the scores by a low temperature and the gap between first and second choice explodes: the top word wins almost always, so the output becomes focused, deterministic, repetitive. Divide by a high temperature and the scores flatten: unlikely words get real chances, the output varies more, and at extreme values it turns to noise. At exactly 0 the dice disappear and the model always picks the top word (greedy decoding). Temperature changes style, not knowledge: the same model at 0 and at 1 knows the same facts, it just commits to them differently. Where it breaks: near-zero produces loops and safe clichés, above 1.5 produces word salad. Set it per task: 0 for extraction and classification, around 0.7 for drafting, 1 or more for brainstorming.
Terms that live nearby.
Large Language Model
A large language model (LLM) is an AI system trained on massive collections of text to predict the next token, which gives it a general ability to write, summarise, translate and reason without being programmed for each task.
ReadToken
A token is the chunk of text a language model actually reads and writes: on average about three quarters of an English word, converted to a number from a fixed vocabulary before the model can process it.
ReadInference
Inference is using a trained model: you send input in, the model applies what training encoded in its parameters, and you get an answer out, one token at a time.
Read§ 03
Want this running in your stack?
We design and ship AI systems with permissions, logs and human approval. It starts with a 30-minute scoping call.