AI-103: Developing AI Apps and Agents on Azure

The temperature parameter controls how much randomness is introduced when the model samples the next token: higher values increase diversity and unpredictability of output, while lower values make the model favor the most probable tokens, producing more deterministic and repeatable responses for the same input prompt. Lowering (decreasing) the temperature therefore reduces response variability across repeated calls with identical prompts, satisfying the automated testing requirement, without requiring any change to the prompt text and without degrading factual accuracy—in fact, low temperature settings are commonly recommended for accuracy-sensitive, deterministic use cases such as RAG and conversational Q&A validation. Changing max_tokens only affects response length limits, and removing stop sequences only affects where generation is cut off; neither controls sampling randomness.