The response is either a hallucination or, more likely, it’s using a system prompt where that is stated.
You cannot “ask” a model about which GPU it’s running on either btw 
LLMs are token factories. They are not usually aware of their own configuration. This will be in the metadata of the model that is inaccessible during inference.
You should engineer assuming you can’t know the cut-off date. Then you can switch models later without relying on this.