THE Chatgpt It can consume up to 500 ml of water every 20 to 50 questions. It is as if the platform drink a bottle of water every time this volume of commands is requested by the user. In more complex demands, such as image generation, water expenditure is larger and faster.
To generate an image, it is as if the OpenAi artificial intelligence platform (IA) performed 20 text commands. However, in the case of texts, the same amount of water could be used for up to 50 commands, which does not occur in the illustrations.
This is what a research points from the universities of Colorado Riverside and Texas Arlington released in late March.
ChatgPT water consumption is not new. Its processors, which work 24 hours a day, require a significant amount of energy and need to use this natural resource in the cooling process.
The discussion has gained new contours with the fashion of recent weeks that has gained thousands of internet users: Ask ChatgPT create images with the Studio Ghibli style and from its own photos .
“It’s superdivered to see people loving the images on chatgpt, but our GPUs are melting. We hope it doesn’t take long. The free version of ChatgPT will soon do three generations a day,” wrote OpenAi SEO, Sam Altman, in X by commenting on the sudden growth of demand.
Nina da Hora Ia expert explains that this problem is not private to OpenAi, but a challenge that permeates all the generative Ias. “Tools like Midjourney, Dall · and and other image generation platforms consume a significant amount of computational resources at each request, and this is directly reflected in the use of water on data centers to keep machines at proper temperatures. This daily use, regardless of current fever [relacionado ao Studio Ghibli], continues to generate this impact “, Analyzes.
The news is now the surprise among users, as the technology industry has always acted as a large consumer of natural resources, due to the processing data they perform, according to the Professor and specialist at IA Fernando Moulin .
“This processing consumes a huge amount of energy, and the passage of this energy through conductors and processors, for example, causes a physical effect that transforms electrical energy into heat,” he explains. “This heat generated by the phenomenon is massive, and there are machines called ‘heat exchangers’, which use cooling water to lower the temperature of the servers and thus prevent them from overheating or deteriorating while operating.”
At the height of the new trend, OpenAi announced that 1 million new users had entered the platform in just one hour and that 3 million images were created in a single day. Based on this information, it is possible to estimate that the Water spent, just from this audience, was approximately 75,000 liters in a 60 -minute window.
The water impact of the Ia
It is estimated that, by 2027, the global demand for AI represents between 4.2 and 6.6 billion cubic meters of water capture – The equivalent of the total annual water withdrawal of four to six Danes or half of the United Kingdom – highlights the survey released in March.
“This is worrying, as freshwater scarcity has become one of the most urgent challenges today,” says scientists who head the study. “To respond to global water challenges, AI can and must take social responsibility, giving an example by addressing its own water footprint.”
For the Professor and Specialist in Ia Adilson Batista the lack of transparency of AI companies prevents further analysis of their carbon and water footprints. “Large technology companies still disclose little data on the water and energy consumption of their models, which makes it difficult to understand the real environmental impact of these tools,” he says.
“To increase consciousness [das informações]it would be important for these companies to adopt disclosure policies [divulgação] clearer, publishing regular reports with environmental indicators. ”
Measures to reduce consumption
The study of the universities of Colorado Riverside and Texas Arlington also suggests some strategies to minimize the water impact of AI models, such as the development of more efficient cooling systems, the choice of data centers in cold regions and the adoption of alternative technologies that reduce the need for water. Regulation is also pointed out as a possible path by researchers.
“Deciding carefully when and where training a large AI model can significantly impact your water footprint,” the study says. “We can dynamically schedule the training and inference of AI consciously regarding water use, reducing its water footprint.”
In addition, the text proposes that users and companies consider the environmental footprint of the digital tools used . As well as the concern with carbon emissions and energy consumption has gained prominence in recent years, the use of water can become a relevant criterion in the adoption of new technologies .
“To reduce carbon footprint, it is preferable to ‘follow the sun’ when solar energy is more abundant. However, to reduce water footprint, it may be more advantageous ‘not to follow the sun’, avoiding high temperature hours of the day, when efficiency in water use is lower,” concludes the research.
See also: Chatgpt earns 1 million new users
This content was originally published in ChatgPT spends a lot of water to produce images; Understand on CNN Brazil.
Source: CNN Brasil
Charles Grill is a tech-savvy writer with over 3 years of experience in the field. He writes on a variety of technology-related topics and has a strong focus on the latest advancements in the industry. He is connected with several online news websites and is currently contributing to a technology-focused platform.
