I am an AI tester who test various outputs of AI application, even I play with ChatGPT to output what I want using various prompts.
While testing the output I always wonder that whatever answers these AI applications are providing they are providing either from the provided data source or from internet, are they violating copyright of any original data?
So I tried to check this fine tuning my system message and I can say I have included correct instructions to application to avoid copyright voilation or infringement.
But the question is how will I make sure it is not doing it as I don’t see major diff in before and after answers with system message fine tuning, what I did is increated my own data source with some information and asking the question around it before adding any instructions to system message and after adding instructions asked the same question but I didn’t see major difference
Basically my main challenge is how I make my application to do copyright infringement first, It is a chatBot application using openAI services.
I request the community, If someone can shed some light on what test data I should use in my document and also how to replicate copyright voilation
Thanks
To check wether the AI tool is using copyrighted material or not without it having knowledge about it would be done by retraining it with an extra label of copyright/not-copyright. Thats an enormous task for chatgpt.
What does the tool say about copyright?
What you could do is check the answer source on internet and see yourself if its copyrighted or not.
1 Like