> do you guys seriously think that OpenAI spends their time scanning chat logs for ideas?

While I think along the same lines, I still imagine that an agent would/could make the perfect score reading everything and prioritize ideas found in the chat logs. I would even think this should be a great way to find some hidden user feedback (like people complaining to the chatbot about xyz idea).

On the second part, yes: memory on/off sounds definitely like a feature almost every user thought about.