Google’s Free Access to AI Raises Concerns Over Climate Impact
A recent study by German researchers has found that large language models (LLMs), available for free access, have a significant environmental impact. The 14 LLMs tested, from top developers like Meta and Alibaba, were asked to answer 1,000 challenging academic questions. Using powerful hardware and specialized software, the researchers measured electricity consumption per answer and converted it into carbon dioxide equivalent emissions.
The results showed that many LLMs are far more powerful than needed for everyday queries, using excessive energy and producing higher greenhouse gas emissions. Even smaller models can answer factual questions just as well, reducing their environmental footprint. The study highlights a trade-off: while accuracy improves with larger model sizes, emissions increase dramatically.
One notable example is the Cogito model, which achieved high accuracy but produced three times more emissions than similar-sized models. This raises questions about the necessity of massive 400-billion parameter GPT models for everyday queries, such as answering historical events like World War II.
Source: https://www.fastcompany.com/91354353/ai-users-have-to-choose-between-accuracy-or-sustainability