Tone speaking to AI

My friend once called out that I am very polite when I speak to Siri. I actually never thought much. I just thought this is something that I should do. The positive impact of good communication is not only for other people, but also for myself. This is why I never questioned to speak politely even to a robot.

However, since GenAI comes to the world, there has been a famous argument - saying “thanks” to AI wastes significant amount of electricity. Even funnier, someone even experimented something called “kitten prompt” (basically convincing the model if it misconducts, a kitten will die), and claimed that it actually improves the output accuracy. And even from a pricing perspective, cutting words like “thanks” surely would save some tokens That made me re-evaluate my tone when I speak to AI.

Eventually, I insisted speaking “thanks” most of the time when I want to, and I concluded it is not always a waste of electricity.

As I have always been thinking, I believe that the communication style that we are used to, to some extent, internalises who we are as a person internally. We externalise those traits again without further processing. I am afraid of getting used to be rude might unconsciously leading to bad manner when I speak to human, and therefore, I don’t fully agree on the point that “thanks” is useless.

From a pure technical point of view, I used to think it’s pointless indeed, until I recently checked my Claude memory. I surprisingly found out Claude understands my preference so well.

I thought about it. I think the feedback that I have given to those GenAI tools might have made a difference. For example, I like seeking feedback, and I tried to asking feedback from ChatGPT once. Later after I tried Claude, I also asked Claude for feedback. By that time, Claude told me that I haven’t enabled memory feature, so it doesn’t know anything beyond that particular conversation. I just checked, and this is what I sent back:

Thank you for being explicit and honest. I appreciate your frank communication for not being able to do so, instead of generating hallucination to please me. I also thank your advice to enable memory feature. I have enabled it, and I will come back to ask the same question again some time later. Thank you!

This might fall into that bucket of being too appreciative to AI, but I think it actually delivers a positive feedback about that behaviour I like. I don’t know if Claude would still figure out my preference somehow even if I don’t interact with it this way, but I assume direct feedback would be an easier path to build an accurate profile.

As much as I say “thanks” to AI, I also don’t just say “thanks”. I try to make sure there is some actual information in every message that I send. After all, I am also aware that I am only interacting with a system. There is no need to end a conversation gracefully, or go back to a confusing conversation to clarify it again (if it doesn’t mess up the memory) once I fully understand what actually happened.

I am not sure if I will forever stick to this belief, but this is what I am doing now, and I am happy with it.