The memory is incapable of differentiating its track of conversational tasks. This is regardless of input type.
I even thought it might be due to the ‘personal information’ walls to ensure DPA breaches, and even when prompted with fictional or book writing initiation starts, the error still occurs.
I have this estimated at two inputs minimum before the memory fails. Suppose the input is to create a suggested best output before the final request. In that case, the context of the input information becomes entirely out of context with no connection to its earlier conversational context.
Ex.
Input: You are a detective. What type of questions are best to ask for x outcome?
Output: Good output regarding x objective.
Input: Here is the subject matter ‘theft in the courtroom’ for you to use, Don’t generate a response. Just let me know if you read and understand.
Output: Good output, yes and understand.
Input: Create questions using the information provided while ensuring your suggestions are implemented.
Output: Did you see an elephant cross the road at 9 am?