Three bottles of water are used for GPT-4 to generate just 100 words

A recent study from the University of California has revealed that the artificial intelligence model GPT-4, OpenAIuses up to three bottles of water to generate just 100 words. This figure has brought into focus the high resource consumption of AI systems, at a time when water shortages are affecting various parts of the world.

The operation of the servers that support artificial intelligence requires a constant process of refrigeration which in most cases is done through systems that use water. This resource is used to cool equipment in data centers, which generate large amounts of heat due to the intensive calculations required to provide answers to users.

The report highlights that this excessive water consumption has significant environmental implications, As data centres have become major consumers of water in the areas where they operate, a simple interaction with an AI chatbot may be inadvertently contributing to the environmental crisis, researchers say.

This information adds to the criticism that OpenAI CEO Sam Altman himself has launched about its flagship product. Altman acknowledged that, despite GPT-4’s advances compared to previous versions, the model is still far from reaching the level that the company hopes for the future. “GPT-4 kind of sucks,” he admitted in an interview, referring to the challenges facing the technology.

In addition to resource consumption, GPT-4 has been criticized for other shortcomings such as generic responses, lack of transparency in sources and copyright issues. All of this has generated a growing debate about the global impact that big technologies are having, both in the legal and environmental spheres.