Ai answers you … but: This article explores the topic in depth.
However,
Ai answers you … For example, but:
The debate on the carbon footprint of AI is not new. Consequently, It is powered by a jungle of often partial, sometimes contradictory figures, and a deafening silence of technological giants. Consequently, How many liters of water consumes a conversation with a chatbot? Similarly, How many Co₂ generates the training of a model capable of writing dissertations or code? Consequently, Until now, the responses have been more of the estimate than science.
It is this status quo that Mistral AI. In addition, in collaboration with the Expert Carbon 4 firm and with the support of ADEME, decided to break. Similarly, Less than two years after its creation, the company reveals the gross figures of its most powerful model, Mistral Large 2. Nevertheless, And the results make it vertigo.
The hidden invoice of intelligence – Ai answers you … but
After 18 months of development. use, Large Mistral ai answers you … but 2 training required:
- 20,400 tonnes of equivalent Co₂or the annual carbon footprint of nearly 2,000 French people.
- 281. 000 cubic meters of waterthe equivalent of 75 Olympic swimming pools (or the average annual consumption of 5,200 French), mainly to cool the thousands of processors (GPU) running at full speed.
- 660 kg of antimony equivalenta unit measuring the exhaustion of rare mining resources necessary for the manufacture of servers.
The study reveals, without much surprise, that the training and use phase concentrates almost all of the climate impact (85.5 % of GHG emissions) and on water (91 %). Conversely, it is the manufacture of computer equipment that weighs the heaviest on the exhaustion of planetary resources.
On a user level, the impact of a simple question posed to the Assistant Le Cat seems minimal: 1.14 gram of Co₂ and 45 milliliters of water For a response of 400 tokens. A “marginal” figure according to Mistral. which underlines that the vast majority of emissions take place long before, during the training phase. But relating to the numbers of daily requests that such a service deals with (Chatgpt processes 2.5 billion requests per day), this marginal calculation becomes colossal. If the study figures are applied to the OpenAi service, we reach 2,850 tonnes of CO₂ emitted every day … About 300 times the average annual carbon footprint of a French person (based on an average of 9.4 ai answers you … but tonnes/year).
A call for “environmental competition”
Beyond its own figures, Mistral launches a powerful call to all industry. The objective is not to self -teller, but to create a global standard. The French startup pleads so that the publication of environmental impacts (training. use) becomes an obligation for all AI developers.
The idea is to allow companies and citizens to make informed choices. By creating a kind of energy label for AI models, we could identify and favor the most “frugal” solutions. “It is therefore important to choose ai answers you … but the right model for the right case of use”hammers the study. stressing that the impact is almost proportional to the size of the model. Using an AI mastodon for a simple task is an ecological aberration.
This approach puts pressure on its competitors. Openai has never published a detailed report on Chatgpt. Google, while publishing a global annual report, recognized that AI made its objective of carbon neutrality much more difficult to achieve.
🟣 To not miss any news on the Geek newspaper, subscribe to Google News and on our WhatsApp. And if you love us, we have a newsletter every morning.
Further reading: UBP sees its net profit fall in the first half – Pedophilia: Can we alert the population after a release? – Ottawa helps the steel sector, heavily assigned by prices of 50% – Actions related to cryptocurrencies progress after the signing by Trump of the Stablecoins Act – Byd’s boss announces the arrival of a third brand of electric cars.
