I am currently working as a prompt engineer a b2b coaching company that helps train employees of other companies on how to deal with customers etc. We are now using AI to play the role of a customer to simulate what an interaction with a customer might be like so that our clients can practice. For example: We use ai to play the role of an unhappy IKEA customer, whose bed fell apart just days after building it. And our clients can practice on this ai and try to answer its complaint via a simulated chat conversation. They are then rated on their interaction by our software so that they can improve their interaction…
So the biggest problem we have at the moment is creating a template of this AI that we can apply to multiple companies. Some of our clients are banks, etc. Currently I have the AI tuned pretty well, but occasionally it forgets what role it is playing and will switch roles and play the customer support role instead of the role of the disgruntled customer… another is the AI doesnt act human enough… for example if you told a human to piss off, he would probably hang up on you, not restate his purpose. I intentionally tried to interact with one of our AIs in a really vulgar way and it eventually even said “That is really unprofessional! i will give you one final chance to help me with the issue of my bed frame breaking” which a human would just never do. So I would just like to see if anyone has any ideas on how this can be improved? And if anyone has any questions where I can be of assistance, please ask