Dear XYZ,

Various people have talked about why it is unnecessary to use courteous language when instructing LLMs. Since an LLM has no feelings, there’s no need to say things like “please” or “thank you”.

Some might argue that using language in specific ways could boost the efficacy of such models, but this is not my concern. Beyond matters of efficacy, which is more concerned about the here and now, I’m more interested in how LLMs change us, how we use language, and our society as a result.

Given that an increasing share of our communication may be to non-human language machines, and the medium of exchange is the same (i.e. whatever language of choice). This medium will start to co-evolve with the use of this technology. This is akin to how SMS caused the proliferation of abbreviations like “lol” or “omg” due to the character constraints of that early cellphone messaging service. Today, these abbreviations are effectively universally understood to most English, and perhaps even non-English speakers.

If I have to make a choice between burning a few more tokens or causing our language interactions to become increasingly imperative or impolite, I’ll choose the former.

I should also caveat that given my day job working in AI infrastructure, the cost of those additional tokens pale in comparison to the full context in most LLM conversations. It’s no big deal for the most part.

xoxo,
K