I try to be polite when I write prompts for LLMs. Especially in languages like Korean or Japanese that have grammatical honorifics, I make sure to use the formal, respectful form of speech (what's known as 敬語—gyeongeo or keigo). I joke with my friends that I'm using polite language early on to be pardoned for my sins when AI eventually takes over the world, but the real reason is that I don't want to get used to speaking to someone in a commanding tone. It makes me think I might start believing it's “okay” to order around certain intelligent beings, almost like condoning slavery.
Discussion
Loading...
@hongminhee you know how they say "it will reflect badly on you"
– well it is a positive feedback loop (strengthening your perceived negative behaviour to behave more "bad"), and it has a negative feedback loop side quest (your continued input is suppressing)
Luckily you can reason your way out of the situation, and learn to be proactive
@hongminhee I'm 100% there with you about showing respect to all intelligent beings; to all living beings and products of nature, really.
But I see LLMs as none of that. Not alive, not intelligent. Their output is not the product of human intellect either; it's a lossy compression of stolen strings.
So I'm making a point of being the opposite of respectful when I formulate their input. This helps me avoid imagining there's anything sentient there, not just weights and activations.