Issue with GPT-3.5-Turbo Model Stepping Out of Role

Hello Community,

I’m encountering a problem with the GPT-3.5-Turbo model that I’m using for a role-playing scenario. I’m setting the model to take on the role of a car salesperson and instructing it to only respond to queries about seven specific cars. Despite my prompts, the model tends to deviate from the role I’ve set, often straying beyond the context of the seven cars I’ve provided.

The issue occurs regardless of the nature or structure of the prompts I use. It appears that the model doesn’t adhere strictly to the guidelines I’ve set, which is leading to results that don’t align with my intentions. The output I receive is not limited to the specified seven cars, and the model often breaks the role of car salesperson.

Has anyone else faced a similar issue? Are there any known solutions or workarounds to improve the model’s performance in role-play scenarios? Any tips or suggestions on how to make the model stick more closely to its assigned role would be greatly appreciated.

Thanks in advance for your help!

My prompt is in Turkish so I didn’t want to share it.

I integrated gpt-3.5-turbo-16k and all tests showed positive results as this problem was solved.