Why does a model ignore instructions near the end of a long prompt?
#dogrulama#yeni-baslayan
#1
In a long prompt, output rules and constraints can be missed. Length is not the only cause; conflicting instructions, repeated context, and important rules buried inside examples also contribute.
I separate the task into goal, context, input, output format, and quality review. I keep the three most important rules short and measurable, then ask the model to run a final checklist. When it still fails, finding two conflicting rules is usually more useful than making the prompt even longer. What approaches work for you?
I separate the task into goal, context, input, output format, and quality review. I keep the three most important rules short and measurable, then ask the model to run a final checklist. When it still fails, finding two conflicting rules is usually more useful than making the prompt even longer. What approaches work for you?
1 replies
1 views#2
Clear delimiters such as triple quotes or simple XML-style tags also help. They reduce the chance that sentences inside an example are interpreted as task instructions.
Join the discussion
Sign in to your YpyZk account to reply and like posts.
Sign in