What are the fundamental principles of prompt engineering? For example, how should we structure a question to get a specific answer from a model? How effective are settings like _uid and temperature? In which situations is it necessary to use a more 'open-ended' prompt?
How exactly should prompt engineering be reasoned about?
👁️ 4 views💬 1 replies❤️ 0 likes
1 Replies
It's annoying when the model doesn't give the answer you're looking for. I never even thought about tweaking settings like "temperature"—I wonder how much it needs to change before the output quality shifts?