A few days ago, OpenAI CEO Sam Altman was responding to netizens on the X network about the cost of saying please and thank you to AI models. “Tens of millions of dollars well spent.”
Looking at AI platforms from just one perspective leaves all others aside and results in partial analyses.
In November 2024, during Hunhu eXPONENTIAL, I had the opportunity to speak about “The Thirst of AI: 1 liter of water to write 200 words“. The perspective of this talk was to analyze AI solutions beyond the discourse of AI as a solution to all the world’s ills. And, mainly, contextualizing sustainability issues, considering the energy demands of AI tools.
The problem of energy consumption and generation remains especially with the large investments by Big Tech in nuclear power plants and modern nuclear reactors.
The issue of using please and thanks in prompts cannot be reduced to being polite or the electricity consumption associated with the characters used. After all, AI models are trained and every interaction counts. Kurt Beavers, director of the Microsoft Copilot design team, says the language we use will set the tone of the response. According to Beavers, when an AI model “detects politeness, it is more likely to be polite in its response.”
On the other hand, going the other way around, choosing a speech with swear words and profanities can be very useful. One nimble way to use Google’s AI to find a human-written answer from a credible and viable news source is to use a prompt with either fuck or fucking in the middle. Google’s AI has no idea what to do with swear words, so it reverts to its default mode of providing links.
We cannot forget the factor of a possible – or fanciful – universal dominance by AI. Falling into conspiracy theory territory, if one day machines take over the planet and we become hostages to their whims, then we have been kind to them so they tend to treat us better, or less badly when that scenario becomes real.
