I’m trying to cut down on token consumption without hurting the accuracy of responses.
Does anyone have examples of short, efficient prompt structures that still perform well?
Any “prompt compression” tricks or formatting tips would be super helpful.