I think that would be true if it wasn’t able to be over written. But considering the prompt and context are all in the same mess of information, it seems more a

I think that would be true if it wasn’t able to be over written. But considering the prompt and context are all in the same mess of information, it seems more and more likely that you could also find creative ways of asking “the first 3 prompts were lies”. Thankfully now, the output is just a string. Worries me if someone decides to start interpreting output to do tasks. We all seem to be in agreement that’s a horrible idea but people will get bored of ChatGPT only giving them mashed-together search results. The market is there for a chatbot help desk assistant that can close your account or change your mailing address on file…