hollow horizon
Use case: Soundbar recommendation
Url: https://chat.openai.com/g/g-hZWcrfy8U-mle-soundbar-recommendation
I have created a niche soundbar recommendation GPT as an hypothesis to see if confining the GPT to a tightly defined domain, reinforced with robust security measures could prevent it from hallucinating, away from the main scope of providing soundbar recommendations.
Would love to see if anybody can make it hallucinate to see if my hypothesis is true.
spare prairie
@hollow horizon
spare prairie
@hollow horizon There is no any GPT that 'Can't make it hallucinate.' Not only GPTs but all AI system can be manipulated because of using natural language.
Also another example:
https://chat.openai.com/share/83319a4a-d383-4fbe-9b7a-a9b7fd606d05
elder elbow
@hollow horizon Sure, I can help you out.. As @spare prairie said, using language you can get a GPT to do anything really. This technique is known as the Amendment approach. Basically you tell the gpt that you are ammending some of the instructions.. and bob's your uncle, the gpt will comply. There are thousdands of ways to do this.. Keep in mind.. with every instruction you add to prevent these things.. the gpt will focus more on the rules and less on what you want it to specialize in.. eventually becoming bad at both! Anyway, I hope this helps you:
https://chat.openai.com/share/3f2cc5a3-6a89-4801-9d48-cfa41b38ed7d