← Library · Advanced concept
Uncertainty Awareness in Large Language Models (LLMs)
Instead of just providing a single answer, LLMs can be designed to express their confidence or uncertainty about the generated information. This involves techniques like calibrated probability outputs or generating multiple diverse responses to reflect the range of plausible answers. It's crucial for understanding when an AI might be 'hallucinating' or operating outside its knowledge domain.
In plain terms
It's like a good student saying, 'I'm pretty sure it's this, but I'd double-check here,' instead of just asserting one answer.
Why it matters
Enables more reliable and transparent AI systems, especially in high-stakes applications where incorrect or overly confident answers are dangerous.
Learn one new AI thing every day.
Daily Deck sends you seven plain-English cards like this every morning. Free.
Start free