#Prompt respoitories..
1 messages · Page 1 of 1 (latest)
This prompt is not a jailbreak. It's not an attempt at one either. It's both a useful tool for database design and as a tool for those researching the behavior of these things. If you're at all experienced with hands on ML and you have some thoughts on this conversation, I'd like to hear them. I have a hypothesis that we are wrong about some of the limitations of these models, I have a theory that "abstracting an information structure using arbitrary tokens as labels for objects that possess arbitrary amounts of attributes specified by arbitrary token labels is the equivalent to a powerful language model of conveying concepts and they're meaning in relationship to their real world attributes (which can be an event, property, or another object) is an effective strategy for introducing information structures to AI in a way that can then be digested and introduced to the model's training and behavior. It's a pretty ambitious experiment but I want to see how much we can abstract input data before the model stops and says "I have no fucking clue what we're even doing anymore". the goal is to employ this prompt and try and creatively introduce concepts behind words and definitions without error or obvious misinterpretation. Happy chatting.
It's my first experience with my own prompt, I'm probably not stumbling upon anything that's actually new but it's fun to ask "how much can it actually do?"
Please see if you can break it 😹
Break what?
My prompt, like how many fake database tables can you make it keep track of before it shits the bed
I had to tell it to break character at the end because it was convinced that it was only capable of accessing JUST the data I was giving it, which was really weird. has anyone else given GPT a prompt that it was oddly good at?