Openai CEO Sam Altman revealed that courtesy to Chatgpt alone could cost “ten millions of dollars” of additional computing resources.
When asked about the huge amount of money Openai lost the cost of electricity for people who said “please” and “thank you” to their AI models, Altman replied: “It cost tens of millions of dollars. You never know.”
Spend tens of millions of dollars, you never know
– Sam Ultraman (@sama) April 16, 2025
Every word sent to chatgpt, even ordinary etiquette – requires additional processing power. even The seemingly small interactions quickly add up, increasing power costs and server usage in billions of conversations
The AI model relies heavily on huge data centers that already account for about 2% of global electricity consumption, which is expected to climb, and AI may drain the same energy as industrialized countries such as Japan.
According to a survey conducted by The Washington Post in collaboration with University of California researchers, sending a 100-word email using AI requires 0.14 kWh of electricity — enough to power 14 LED lights for an hour. On a large scale, these small interactions produce a huge energy footprint.
Water is also eye-catching. UC Riverside researchers found that using GPT-4 produces 100 words, consuming up to 3 bottles of water to cool the server, and even a simple three-word response like “Welcome” uses about 1.5 ounces of water.
Why are users polite to machines?
There is evidence that people are really trying to practice sound AI etiquette. A survey in late 2023 found that 67% of U.S. respondents reported being good at their chatbots.
Of those who practice digital politeness, 55% said “doing this because it’s the right thing to do”, while the more cautious 12% admitted that doing it was to “please the algorithm in the case of an AI uprising.” AI has come a long way since 2023, and we have seen a lot of doomsday theories make headlines, so the number is probably much higher now!
Instead of AI etiquette being purely a psychological or behavioral problem, A study by Waseda University researchers found that using polite tips can actually produce higher quality responses from large language models (LLMs).
After all, LLM is trained in human interaction; “LLM reflects a desire for humans to be respected to a certain extent,” the researchers explained.
Human Elements
In addition to technical performance, some experts believe that it is valuable to maintain politeness to AI for our own benefit. youSinging disrespectful language with AI may normalize rudeness in our human interactions.
This phenomenon has been observed by early voice assistants. Some parents reported that their children became more respectful after getting used to Siri or Alexa’s barking commands, and he led Google to introduce their “Pretty Please” feature to encourage 2018 child courtesy.
So whether you are doing the best practice for performance reasons, or keeping the right side of AI in a matrix takeover, whether you are AI-friendly or staying on the right side of AI, just realize that each interaction will bring costs.
