Please, Thank You, and Tens of Millions of Dollars Later…
Photo by Dmytro Bayer @dmytrobayer on Unsplash
I knew it. Manners are a thing of the past. You give a child the last few bites of food from your plate, and they receive it by ramming it in their mouth.
Without a word of thanks
The little rascals. Now they’re being trained to avoid the lovely words of please and thank you.
By Sam Altman.
Get On With It. Now.
There isn’t anything wrong with brevity, the fine art of getting straight to the point. Programmers are often obsessed with the idea that having shorter, concise code is a goal in itself (code golf being an outcome of this mindset).
So, perhaps it’s no surprise that OpenAI CEO is kicking back against people adding pleasantries to their prompts. They say that saying “please” and “thank you” to ChatGPT is expensive. As in, “tens of millions of dollars” expensive.
Yes, really.
It’s Going Wrong
It’s a quaint little habit. You might be writing please and thank you out of courtesy, guilt, misplaced optimism or superstition that the killer robots will spare those that treated our AI overlords with respect.
Our parents taught us that politeness costs nothing, so they might be surprised that those three little words are burning through the electricity that would light a medium-sized town. Those that thought we were wasting the earth’s natural resources with a plane journey for a child to see their parents need a realignment.
The Politeness Tax
Those extra words (tokens?) being thrown at large language models that GPUs must chew through, powered by data centers guzzling enough kilowatts to make Greta Thunberg say “how dare you” to your face.
“Losses well spent,” Altman quipped, which is Silicon Valley code for “we have no idea what’s happening anymore, but it looks cool, so we’re doing it anyway.” The statement “you never know” was tacked on like a post-it note of philosophical doom, or possibly a line from a Black Mirror episode.
Let’s break this down: adding “please” and “thank you” to your prompt means the model has to parse and interpret more words, which means more GPU cycles, which means more power, which means more carbon dioxide in the atmosphere, which means polar bears floating away on ever-smaller icebergs. All because we didn’t want to sound like jerks to an algorithm that couldn’t care less. Is that what the future will really look like?
AI Etiquette
At work, someone proudly made a demo where they solved a simple bug. They claimed they were vibe coding and using AI to solve a problem that could have been quickly solved with a human brain (and human experience gained).
We can’t be surprised that the defenders of AI etiquette are lining up to tell us that this is good for civilization. A Microsoft Copilot designer insisted that polite prompts “help generate respectful, collaborative outputs” because, I guess, nothing screams collaboration like being gaslit by Clippy.
A Microsoft WorkLab article claims AI is more likely to be polite when it recognizes politeness. Because nothing says objective machine logic like “vibes-based responsiveness”. I guess the people who name their Roombas, swear at Siri and think Google Maps is possessed by a directionless goon are taking over the world. Those people are costing OpenAI the GDP of a small country in server fees.
Conclusion
You know what. When Skynet boots up, I’m going to be one of the remaining humans.
Because I think it’s going to remember that I said please when I instructed it to write a thank you note.
Yes, I’m the reason the earth is heating up. Yes, I’m the problem. I guess someone is going to instruct Skynet to target me first?