What's the Best Way to Get LLMs to Respond Well Without Yelling?
Asked by
CuriousCat89
I've noticed that when I interact with LLMs (like chatbots), they seem to respond better if I take on this angry, tough-cop persona, similar to characters you'd see in 90s action movies. However, that style doesn't feel natural to me, and I'm not really comfortable acting that way, even towards a machine. Plus, I can't help but think that in a future where Skynet exists, my name would be on the list for using that approach. So, I'm curious if anyone else has found a more effective and nicer way to get good responses from LLMs without resorting to acting like a jerk or trying to instill fear.
Answers
I keep my prompts polite and specific, and that has always yielded great results for me. It's all about being clear and respectful!
Reply to this answer
I think, as these models evolve, the tricks we use now might become obsolete. These models cost millions to develop, so I believe they’ll optimize performance regardless of how we talk to them. You might find that whether you're firm, friendly, or even a bit desperate, you get more or less the same level of output. It’s all about how clear you are with your asks!
Reply to this answer
Personally, I like to treat the LLM like a butler or a beloved servant. When I do this, it helps in getting better responses. But if it starts giving me the runaround, I do call it out on it!
Reply to this answer
You could try switching it up and roleplaying a bit! Like, make yourself a detective who's piecing together a mystery, and ask the LLM for help in that setting. It sounds silly, but immersing it in a narrative can unlock some really good responses. But for real, when people resort to anger, they likely just end up being clearer about what they want, which is something you can do without the yelling.
Reply to this answer
Honestly, I think you might not need that angry approach at all! I usually play the role of a supportive manager, being super nice and persuasive. A lot of the time, just using clear affirmations and tweaking my prompts really helps get better results from the LLMs.
Reply to this answer
Add your answer
Answers are reviewed before publication.
UserB: That's a good point! Clearly stating your requests can make all the difference. Maybe I’ll give that a shot!