Question

A marketing team uses an LLM for two separate tasks: (1) generating varied, imaginative slogan ideas for a brainstorm, and (2) producing standardized product descriptions that must stay consistent and factual across a catalog. A team member asks how a single tunable setting could help balance these different needs. Which setting most directly controls the randomness and variability of the model's generated output?

Accepted Answer

The temperature setting, which adjusts how random or deterministic the model's word choices are

Answer

The context window size, which sets how much prior text the model can reference at once

Answer

The token limit, which caps the total length of the generated response

Answer

The embedding dimension, which defines how meaning is stored numerically