Experiences with complex instructions
Hey! Thank you so much for providing this OSS LLM 💙 I've been testing out this model with various prompts and I'd like to see if there are others out there who may share ideas or who've had similar experiences.
As far as I can tell, the best way to prompt this model is to follow a simple pattern:
The following is an interaction between an assistant and a customer.
Customer: Hey
Assistant:
Or
Answer the following question
Question: Who am I?
Answer
However, the moment I add some sort of formatting instructions into the mix, these seem to get ignored. Some examples:"cite the Document[number] which your answer was generated from" (what I'm actually trying) but also simple ones like "add YO at the end of each sentence" or "End your answers with XYZ"
Has anyone had experiences with these types of more complex prompts and if yes, what have your learnings been? Would love to hear
Answer the following question
Question: Who am I?
Answer