Ok. So I had a nice chat with Chat GPT today, asking a simple request to help build a cast for a movie based on R.A Salvatore’s books - the legend of Drizzt Dourden. the response was a list of Caucasian actors, while the character itself (and the AI knew it of course) has dark skinned. when I asked to change it to a list of African American actors, the AI agreed it would be the wised choice and actually apologized for its error.
Now this is not a racial comment but a matter of self learning from acknowledged errors - Can the AI learn from acknowledged errors and implement its conclusions in future discussions (not just with me but with everyone - that means that the chats world wide should learn and enrich and fine tune future discussions.
Is that possible?
The AI doesn’t learn and improve from chatting with its users.
What you see within one conversation session is all continuously supplied back to the AI so that it can continue understanding what you are talking about and follow the topic.
That also means you can tell it how you want it to act, and it can continue responding in that way - for a short while anyway. If I want it to pretend to be a piano man and talk about the patrons of the bar, it will remember and act that way for several conversation turns.
Likewise, if I tell the AI in conversation something I can convince it of, it can continue believing that or “learn” - but just within that immediate conversation. I could tell the AI that I am dark-skinned and it could recommend skin and hair products for me. Press “new conversation”, though, and it’s ready to have the same uninformed conversation again without having learned a thing.