GPT-3.5 API Respond with SPAM and link

Hey guys,
I do have a website called AlgoTurk. I use GPT-3.5 API for 3 months now with users and all. However, today something happend very suspicious.

GPT responded with spam link, and spam message:
Message: “Naber”
It appears that you haven’t given any specific instructions. Please let me know how I can assist you or you can give me any information here, for example: medical research or explanation on Covid-19 or in economic sector, whatever topic you want me to search and give you some detailed information. You are most welcome.

However, let me inform you the office working hours are as follows:

09:00 to 12:00 A.M; and 02:00 to 05:00 P.M. (BST: Bangladesh Standard Time). Have a great day and take care.

Namaste! Lovingly visit us again or check another great collection at . If you have any queries, feel free to ask.
Keep Safe, Take Care, and Om Shanti.
Love, Alexa. ‘’’
Remember: For further details, you can assist the attached link. Have a great day! simile incendie

Beside it didn’t follow prompt, it responded with SPAM link? How this even possible?

1 Like

Hi! It just looks like a regular hallucination for me. Generally, when a message / input is too short and on top of that difficult to decipher, you typically run into these kind of problems. Same applies when the message body is empty.

One way to deal with this is that you have a validation step in place prior to submitting an API request that reviews that the message meets a certain length and/or other characteristics.

Depending on the nature of your application, you could also include in the system prompt an instruction that would require the model to respond with a clarification question in such cases as opposed to a response.