【question】About benchmark comparison

[wangzhen38]

In benchmark experiments, the test set contains data from multiple domains. Will this lead to the recall of data from other incorrect domains during queries, or do you also consider this as a kind of capability test?

[liukidar]

Hello, yes, indeed when you recall data given a query, it is possible to touch multiple domains, but as the RAG quality improves, this becomes less and less likely