The kindness with chatgpt is coming out of OpenAi. The “please” and “thanks” have an absurd cost every month

Probably these days, doing Scroll In social networks, you have crossed with one of those memes that say everything with an image: a man about to be eliminated by robots is saved at the last moment because, according to one of them, he “always said thanks.” A minimum, almost automatic gesture, which redeems it to machines in the dystopic future and that starts a smile in the meantime algorithm.

Obviously, we do not expect robots to take control of the planet. But it is no accident that this idea has become viral. Cinema and literature have been feeding that imaginary between fascination and fear. The funny thing is that, beyond fiction, there is something real in all this: being educated with artificial intelligence (AI) is not free. Our kindness, however routine, costs money. And not precisely little.

The functioning of AI is not exactly cheap. When we use an artificial intelligence chatbot, we rarely think about the consumption of resources that each of our requests implies. Chatgpt and other similar systems work thanks to large language models, also known as LLM, which operate in data centers equipped with high performance hardware, such as NVIDIA H100 GPU. The execution of these models, which is technically called inference, requires enormous processing capacity and, therefore, a considerable amount of electricity.

To all this we must add the necessary investment to build and maintain these infrastructure, as well as their environmental impact. We know, for example, that many of these centers consume large volumes of water to cool the systems, since this type of computer generates heat. Very hot. There is no consensus on how much each request costs, and the figures vary according to the company. Even so, Sam Altman has given a track.

The amazing cost of kindness. A few days ago, a user asked on social networks how much money Openai would have lost just because people add “please” and “thank you” to interact with their models. CEO’s response did not go unnoticed: “Dozens of millions of dollars well spent.” He did not speak of a single interaction, of course, but about the accumulation of millions of friendly messages over time.


Altman Amability 1
Altman Amability 1

Click to see the original message in x

It is curious that, despite the high energy and economic cost that this represents, Altman considers that it is a justified expense.

Should we be kind to an AI? As in any other area, there are people who prefer to go directly to the point and others who choose to maintain a more careful and educated tone. This difference probably also has to do with the way each one communicates in their day to day. And in this context, kindness is not just a matter of manners.

A study entitled “Should We respect LLMS? A Cross-Linguistic Study of Prompt Politeness in Academic Performance”Analyzed the role of courtesy in the indications and left some interesting conclusions. Among them, that unfortunate messages tend to generate a worse performance. Although, yes, excess courtesy does not guarantee better results.

What seems clear to those who use this type of chatbots daily, and more now that they are able to learn and remember more precision, they can adapt to the user’s tone. If we want a more respectful, more coherent and a little more human interaction, we should probably start by taking that step.

Images | X screen capture

In Xataka | The hallucinations are still the Achilles heel of the AI: the latest OpenAI models invent more of the account

Leave your vote

Leave a Comment

GIPHY App Key not set. Please check settings

Log In

Forgot password?

Forgot password?

Enter your account data and we will send you a link to reset your password.

Your password reset link appears to be invalid or expired.

Log in

Privacy Policy

Add to Collection

No Collections

Here you'll find all collections you've created before.