"Misused"? Strange way to spell "mistrusted'"
> Any responsible deployment of AI technologies requires rigorous assessments to ensure the data and outputs cannot be misused.
Misused, as in taking the LLMs results at face value!
Interesting new problem is added to the mix: how do you verify that the data Copilot presents you is the *complete* set of "relevant* data? With no little additions?
Whilst we are here, how stable are the results of your Copilot enquiry? Is the work repeatable or will an innocuous change in the way the query is worded going to cause a different dataset to be returned?