Google’s claim that a typical query to its Gemini AI consumes only “five drops of water” has drawn criticism from AI experts. Shaolei Ren, a researcher at UC Riverside and co-author of the report “Making AI Less ‘Thirsty’,” argues that this figure overlooks significant factors. He estimates that training OpenAI’s GPT-3 model required 700,000 liters of water, with a ChatGPT conversation consuming nearly a pint. Experts assert that Google’s estimate only accounts for direct water usage in data center cooling, omitting the indirect consumption from power plants supplying electricity. Sustainability researcher Alex de Vries-Gao emphasizes that Google’s figure is merely “the tip of the iceberg.” Additionally, criticisms extend to Google’s carbon emissions figures, which allegedly use market-based metrics, failing to accurately reflect the true CO2 output of AI models. These insights challenge Google’s portrayal of its AI’s environmental impact, raising questions about sustainability and resource consumption in the tech industry.
Source link
Share
Read more