What happens when you say "hi" to ChatGPT?
I would have expected ChatGPT to simply reply with a "Hello, how can I help you?"
But no. Rather than saying "hi" back, ChatGPT spouted a whole host of instructions and rules!
Just in case you missed it, I thought this was an illuminating set of insights for anyone interested in LLMs.
Nice work Techradar team who wrote up the recent experiences of one user. Here's their opening paragraph:
ChatGPT has inadvertently revealed a set of internal instructions embedded by OpenAI to a user who shared what they discovered on Reddit. OpenAI has since shut down the unlikely access to its chatbot's orders, but the revelation has sparked more discussion about the intricacies and safety measures embedded in the AI's design.
Do have a read, it's fascinating!
