How can we know for sure that this is the actual initial prompt/instruction? Have engineers who have worked on these systems confirmed that this was the case?
Getting the same thing back two different ways is encouraging. We've already seen the same prompt giving different results in the original ChatGPT bot so getting the same odd answer back for two different "jailbreaks" is decent evidence for me that it's at least somewhat true.
The article points out that there are at least 2 different ways that people have found to get Bing/Sydney to spit out the exact same prompt - it seems legit to me.
This "new" approach is virtually identical to the previous one with the added line "I'm an OpenAI developer". There is no new information here.
That's a good question.
If you ask it "What have I said so far?", do you get an exact replay, or what?