Being Polite To ChatGPT Costs OpenAI Tens of Millions
OpenAI CEO Sam Altman revealed (https://x.com/sama/status/1912646035979239430) on X that polite user prompts to ChatGPT cost the company tens of millions of dollars. "Well spent—you never know [what might happen]," he added, hinting at a future where rudeness to AI might backfire.
While Altman's tone was ironic and not meant to provide exact figures, his comment sparked a debate: should we waste resources being polite to a machine?
Where Do the Costs Come From?
Each prompt to a chatbot triggers a chain of computations: chips spinning up, cooling systems running, and other infrastructure working in data centers. Even a few "extra" tokens in a prompt increase server load and power consumption.
Researchers at Epoch AI estimated (https://epoch.ai/gradient-updates/how-much-energy-does-chatgpt-use) that a single GPT-4o query requires about 0.3 watt-hours of electricity—enough to boil a teaspoon of water. Words like "please" and "thank you" are a small part of a conversation. Still, across billions of interactions, they add up to megawatts of power—and serious operating costs.
Who Says "Thanks" to AI?
In the U.S., 67% of people who use AI are polite to it. (https://www.techradar.com/computing/artificial-intelligence/are-you-polite-to-chatgpt-heres-where-you-rank-among-ai-chatbot-users) Of those, 55% say it's simply "the right thing to do," and 12% say it's just in case AI or robots ever become conscious and remember who treated them with respect. Some Reddit and X users say it plainly feels wrong to be rude, even to an algorithm.
And maybe they're onto something: research shows (https://aclanthology.org/2024.sicon-1.2/) that politeness can improve AI responses. Curtis Beavers, a director on the design team for Microsoft Copilot, explains (https://www.microsoft.com/en-us/worklab/why-using-a-polite-tone-with-ai-matters) that LLMs mirror the user's tone, responding not only politely but also more professionally. We covered this in more detail here. (https://t.me/hiaimediaen/832)
So, are you polite to AI?