API output seems to be infected by old request

Hi Together,
I have a strange behaviour with my api. My output seems to be infected by an older request.
Some days ago I used it to create multiple choice questions for given topics and it worked fine.
And now I want it to create informational paragraphs about the same topics.
For creating this I use a similar python code in a new colab notebook and same csv file with a list of topics.
However, the system prompts are completely different and still it sometimes creates multiple choice questions instead of paragraphs. Or paragraphs about multiple choice questions.

Maybe someone knows how I can fix this behaviour.

My current prompt is as follows:

def generate_system_prompt(subchapter, chapter):
static_prompt = “You will help me generate text paragraphs.\n”
static_prompt = “I will provide you with a chapter and subchapter of the book Introduction of information systems.\n”
static_prompt += “Use this information as a topic guideline and focus on the subchapter.\n”
static_prompt += “Generate 4 individual paragraphs that cover the most important parts of the subchapter.\n”
static_prompt += “Don’t mention the subchapter.\n”
static_prompt += “Don’t generate an introduction or a conclusion paragraph.\n”
static_prompt += “Divide each paragraph with an empty line.\n”
static_prompt += “Stay below 300 words.\n”

dynamic_prompt = f"{chapter}, {subchapter}\n\nMCQ:"

You’re missing a + before the = on the second line, so it skips the first part (“You will help me generate text paragraphs.\n”), it might not be the main problem but it won’t help xD

2 Likes

Thanks for the reply and you are right, this was missing!
Unsurprisingly it is still creating MCQ questions :smiling_face_with_tear: