Four Tokens of Humanity: The Debate Over AI Etiquette vs. "Efficiency"

Image source: https://joeingermany.wordpress.com/2011/04/06/british-politeness-vs-german-directness/

Summary

People are actually arguing we should drop 'please' and 'thank you' from AI prompts to save money and tokens. Really? We've got mountains of pressing AI issues to tackle, and we're fixated on four tokens? Let's take a step back and re-prioritize what actually matters in the AI space.

Sam Strikes Again

Recently, Sam Altman responded on Twitter (X) to a user asking the following:

“I wonder how much money OpenAI has lost in electricity costs from people saying ‘please’ and ‘thank you’ to their models.” - @tomieinlove on Twitter/X

This has prompted a tirade of posts across the ‘AI Influencer’ space discussing whether or not you should say “please” or “thank you”. A handful of samples below:

• Damie B. - “The short answer, my take is yes.”

• Justin Seeley - “Are we being kind… or careless?”

• Dave Diamond - “You’re training ChatGPT to suck. One ‘please’ at a time.”

So… should we be polite to AI, or not? My take: who gives a shit?

Four tokens? Really?

We have more important things to worry about in the context of AI, frankly.

It’s four tokens according to OpenAI’s online tokenizer to say “please and thank you.” Six if you include the quote marks. Likely closer to 3 if you say “Please” in your initial prompt and “Thank you!” at the end of it. The system prompt for ChatGPT is 1700+ tokens long. Additionally, the average prompt to an AI system is <X> in length. I can understand why people might be concerned in the context of ChatGPT’s millions of users, but ChatGPT isn’t the only available AI tool. Especially for those of us who have our models running locally… does it really matter?

Do we have bigger fish to fry?

To remind you, we still don’t have clear answers and a plan for a number of key issues plaguing wide-scale AI implementation and usage. Naming a few:

• The Environmental Problem(s): Power requirements for data centers have grown from 2,688 megawatts to 5,341 megawatts in 2023. 7% of Signapore’s entire energy usage is specifically used by data centers. That’s not even discussing water or precious metal usage for supporting data center construction and maintenance.

• The Business Problem(s): PMI states that 70% of business projects with AI fail. Companies like OpenAI are by and large not profitable. Customers aren’t even all that excited about AI being added to all of their products.

• The Societal Problem(s): Studies find link between AI usage and decreased cognitive function. 41% of businesses are reducing their workforce by 2030 due to AI. As AI is being adopted in spaces like education, 80% of parents had concerns about harmful effects of AI in education.

That’s just what I could find in a short period of time, and I’m sure there’s countless more ways to slice the problems we’re facing through AI. I find it somewhat disheartening that we’re focused on “please” and “thank you” over any of these topics.

How about the people around you?

Finally, from a personal perspective, I’m worried that people are spending more time focused about how to prompt, how to speak, how to optimize for artificial intelligence work… and not enough focusing on being effective communicators and good neighbors to the humans around you.

86% of employees and executives cite the lack of effective collaboration and communication as the main causes of workplace failures. Conversely, a third of communicators claim to be dissatisfied with the channels they use in the workplace in 2025. Looking a bit higher level to general “vibes,” almost 50% of Americans say that people have gotten more rude since the COVID-19 pandemic. Does focusing on “please” and “thank you” in the prompt you’re giving an AI make you more polite to your coworkers, your neighbors, your friends, your family?

AI can certainly be a differentiating factor to improving communication between humans, though! From the same source I cited above regarding lack of communication, 73% of knowledge workers say that gen AI tools have helped them avoid miscommunication at work. I bet you good money that either the input or output in these generative AI tools included “please” and “thank you,” especially if there’s an interpersonal issue being “navigated” with AI. Is it a waste of resources, a waste of time, a waste of money to be polite?

So where do we go from here?

It just feels like the whole argument of “we’re wasting millions of dollars“ is exhausting to field in the context of basic decency. Even from a budget perspective, Adam Holter cited that the cost of “being nice” is a 0.01% drop in the bucket for how much AI inference costs overall. Sam Altman even said that the “tens of millions of dollars” were well spent. Why is everybody so hung up on this?? We have bigger fish to fry, and we should be focused on bigger problems regarding AI, not the four-ish tokens that adding those statements takes up.

I’m not a doctor, but here’s my prescription:

Be kind to one another 

Talk with your neighbor, get some lunch with somebody, have those tough conversations. Take the time to be nice to the humans around you. Equally or even nicer than how you speak to your ChatGPT window (or LLM model and chat method of choice, I personally use Gemini, Claude, and Qwen2.5 through LM Studio).

Touch some grass

I’d recommend being wary of anybody who is advocating for removing common decency for what is largely a minimal reduction in compute cost. Don’t let anybody try to stop you from being kind. Also: don’t be afraid to ask tough questions about AI on things like the environment, social impacts, and business if that’s your cup of tea.

Recognize that optimization can go overboard

A bit of a hot take in the tech space: There is an optimal answer to every problem. There’s also a wide valley between a set of “wrong answers” and the optimal I love to refer to as “good enough.”

In some spaces, “good enough” is not enough like healthcare, aviation, and other highly regulated industries. Even in customer service, “good enough” can be an excuse for laziness and apathy when used wrong, that of which is recipe for disaster. Beyond those areas? There’s a point where software can only do so much. The convenience of AI doesn’t get more convenient just because it’s faster, or because we’re saving a fraction of the total token count.

I’m not advocating for people to not have standards and not strive for excellence. I just find that excellence, like so many other things, is a spectrum. If your AI-generated marketing plan takes a couple of minutes to be created, that’s time for coffee or tea! Maybe you can even take the eye break from the computer screen everyone says you should do.

The possibilities are endless!