Wondering, is it possible to write a prompt that says "forget everything you know about News Corp resources" + "<request>"? Similarly, could one query "according to New York Post" + "<request>"? If so, that'd be kinda great.
Sure you can, and it'll claim to be doing so, which of course has no bearing at all on whether that's actually what it will do (or is even capable of doing).
Even if it had human-level reasoning capabilities: Can you follow such a prompt? Neural network weights aren't a database.
The latter would be possible, just not with the model alone (ChatGPT/derivatives with web access should do this). For the former, looking on the bright side it’ll give GPT “personality”. Of Rupert, but personality!