AIExplainer

What is unsupported-claim rate?

The rate at which a language model produces responses that are not supported by evidence or facts

The unsupported-claim rate measures how often a language model generates claims or statements that are not backed up by reliable sources or data, which can lead to the spread of misinformation

Think of the unsupported-claim rate like a fact-checker's red flag - it highlights when a language model is making claims that are not grounded in reality, much like a journalist fact-checking a source

For instance, a language model with a high unsupported-claim rate might claim that a certain medical treatment is effective without citing any scientific studies to support the claim

The unsupported-claim rate is used to evaluate the performance and reliability of language models, helping developers to identify areas for improvement and reduce the spread of misinformation

One common misconception is that a low unsupported-claim rate guarantees the accuracy of a language model's responses - however, it's still possible for a model to produce well-supported but incorrect information

The concept of unsupported-claim rate has become increasingly important as language models have become more prevalent and powerful, with many researchers and developers working to improve the accuracy and reliability of these models

misinformation rate factuality rate evidence-based accuracy

Three products for different needs — explore what’s relevant to you.