Well yeah, its using the same dataset as MS copilot.
Spitting out inaccurate (I wish the media would stop feeding into calling it something that sounds less bad like hallucinations) answers is nothing something that will go away until the LLM gains the ability to decern context.