CyberCertPrepAI Security FundamentalsAI & ML Concepts for SecurityEASYFree question
A developer sets temperature=0 for an LLM used in a security alert triage system that must consistently classify alerts into predefined categories. What is the effect of temperature=0?
- A.A. The model refuses to generate any output - temperature=0 is an error state
- B.C. The model generates extremely creative and varied responses, maximizing diversity
- C.B. The model selects the highest-probability token at each step (greedy decoding), producing deterministic and maximally consistent outputs with no randomness
- D.D. The model runs 10x faster because it skips the sampling step
Why C is correct
Temperature controls the randomness of token sampling. At temperature=0, the model always selects the single most likely next token (greedy decoding), producing deterministic outputs - identical prompts yield identical responses. This is ideal for classification tasks where consistency is critical (alert triage, structured data extraction). Higher temperatures (0.7-1.0) introduce randomness, improving variety for creative tasks but reducing reliability for structured tasks. Temperature affects output diversity, not inference speed.
Know someone studying for AI Security Fundamentals? Send them this one.