Temperature changes variation, not factuality
Turning the temperature down makes a model more predictable. It does not make it more accurate.
Low temperature means it picks the most likely continuation more consistently, so you get the same answer each time. If that answer is wrong, you now get the same wrong answer reliably. It is a control over variation, frequently misunderstood as a control over truthfulness, and the confusion leads people to believe they have tuned away hallucination.
More on AI and LLMs
- An LLM predicts plausible continuations, not verified truthIt only checks the shape
- Context is temporary working material, not permanent knowledgeThe board gets wiped
- Retrieval adds documents, not guaranteed correctnessThe filter slot is empty
- System prompts are instructions, not a security boundaryA sign, with no fence
- LLM output is untrusted input downstreamIt comes in round the back
- Fine-tuning changes behaviour, not every limitationNew type, same roller