I try to summarize conversations with clients so to get take aways and to do that I can forward to my clients. But it seems to mix info out of previous conversations and adds them to the new summary.
Even if I state ‘not to use info out of previous conversations’?
What should I write in my prompt to prevent this?
Welcome to the forum!
Start a new conversation for each client. Or even better use a project for each client.
Eric, I did start a new converstation for this client. But GPT took parts of a previous prompt, about the same client, and mixed it up. Even when I stated ‘Don’t use info out of older promts’…
LLMs such as ChatGPT do not understand negative logic, e.g. Don't. What you can do is use words like avoid. However the reasoning capability in o1 models has at times correctly understood negative logic.
You might also check what ChatGPT has recorded in memory.
OK that I understand. That’s like my wife and business partner, she’s the coach and I’m the marketeer, she always tells me and other the same thing when it comes to kids. They don’t understand/comprehend negatives. So we try not to say to our kids what they can’t do but tell them what is available for them to enjoy.
I will use your logic in my next promts. Thanks