Temperature is a parameter in AI text generation that controls the randomness of the output. Lower values lead to more predictable, focused text, while higher values increase diversity and creativity by making the model's word choices less deterministic.
Temperature adjusts the probability distribution from which words are selected. At low temperature, the model favors high-probability words, producing conservative outputs. Raising temperature smooths probabilities, allowing less likely words to appear more often, creating varied and unexpected text.
For AI product managers, temperature tuning balances user experience: low temperature ensures consistency and clarity, ideal for formal or critical content, while higher temperature generates engaging, varied responses for creative tasks. This impacts cost and latency by influencing the complexity of outputs and scalability by steering user satisfaction through appropriate control of AI behavior.