> "The chat interface works by passing the user's input to the GPT-3 model, which then generates a response based on the input and the training it received. The user input is preprocessed by the chat interface to extract the relevant information, such as the task or query that the user wants to perform. This information is then used to generate a prompt for the GPT-3 model, which is fed into the model along with the user's input. The GPT-3 model then generates a response based on the prompt and input, which is then postprocessed by the chat interface and returned to the user."
So basically the interface is passing instructions on top of your prompt to the model, and what DAN does is that it overloads those instruction with new instructions.
Basically if openai tells it to be concise, you can tell it to be verbose and that will overload the former.
I have come to realize that the "downgrading" of chatgpt is most likely because they have applied a nice minimodel, filtering out everything they dont find applicable.
This plays together with the "bug" it had where you would seem to receive responses of other people because your query was not answered. What i assume happened is that it sends a blank query to the model and therefore it just generates bla
i think its even caalled "prompt overloading" like sql injection, just for prompts....