A tiny tweak to AI’s massive language models can drastically cut their energy appetite, say researchers at UNESCO and UCL.
Small changes have a big impact
Large Language Models (LLMs) have become a cornerstone of modern AI. These massive neural networks can process and generate vast amounts of text, from chatbots to language translation software. However, their enormous computational requirements have also made them a significant contributor to greenhouse gas emissions.
According to a new report by UNESCO and UCL, a few simple changes to the way LLMs are built and used can reduce their energy consumption by up to 70%. These tweaks include adjusting the model’s architecture, optimizing the training process, and using more efficient computing hardware.
UNESCO’s mission to make AI more sustainable
UNESCO has a long history of promoting sustainable development in its 194 Member States. In the context of AI, this means encouraging researchers to build more energy-efficient models that can be used to benefit humanity without destroying the planet.
Dr. Marie-Christine Rousset, lead author of the report, emphasizes the importance of responsible AI development. “We’re not just talking about saving energy for its own sake,” she says. “We’re talking about creating AI systems that are more accessible, more equitable, and more sustainable for all.”
What this means
The impact of this research could be significant. By reducing the energy requirements of LLMs, developers can make AI more accessible to organizations and individuals in developing countries where energy infrastructure is limited. This, in turn, can help bridge the digital divide and promote more inclusive and equitable access to information.
