Temperature (Sampling)
A parameter controlling the "creativity" of LLM outputs: Low values (0-0.3) produce focused, deterministic responses; high values (0.7-1.0) bring variation and surprises.
For marketing content: T=0.2 for consistent, fact-based texts (product info). T=0.7 for creative ad copy and brainstorming. T=0.9+ for wild ideation.
Explanation
Temperature scales the logits before the softmax function: T=0 always chooses the most likely token (deterministic), T>1 "flattens" the distribution making unlikely tokens more probable. T=0.7 is often the sweet spot for creative tasks.
Marketing Relevance
For marketing content: T=0.2 for consistent, fact-based texts (product info). T=0.7 for creative ad copy and brainstorming. T=0.9+ for wild ideation. Wrong temperature ruins results.
Example
A team tests headlines at different temperatures: T=0.2 always gives the same solid headline. T=0.7 generates 5 different creative options. T=1.0 produces unconventional, sometimes brilliant, sometimes absurd ideas.
Common Pitfalls
Too high temperature = incoherent output. Too low = boring and repetitive. Optimal varies by task. Interacts with other parameters (top_p, top_k).
Origin & History
Temperature (Sampling) is an established concept in the field of Artificial Intelligence. The concept has evolved alongside the growing importance of AI and data-driven methods.