temperature

When an AI model generates text, it doesn't just pick the single most likely word next. It actually calculates probabilities for many possible words that could come next. Temperature is a setting that influences how much the model explores these other, less likely words. Think of it like adjusting a dial between 'safe and predictable' and 'daring and innovative'.

A low temperature, like 0.1 or 0.2, makes the model very confident and deterministic. It will usually pick the most probable word, resulting in responses that are logical, fact-oriented, and less prone to 'hallucinations' or creative tangents. This is great for tasks where accuracy and consistency are key, like summarizing information or answering direct questions.

Conversely, a high temperature, perhaps 0.8 or 1.0, encourages the model to take more risks. It will consider words with lower probabilities, leading to more diverse, surprising, and sometimes even poetic or imaginative outputs. This higher randomness is useful for creative writing, brainstorming, or generating multiple variations of a text where exact repetition isn't desired. However, it also increases the chance of less coherent or factually incorrect responses.