Li, Pengfei; Yang, Jianyi; Islam, Mohammad A.; and Ren, Shaolei. ‘Making AI Less “Thirsty”: Uncovering and Addressing the Secret Water Footprint of AI Models.’ arXiv, 2023. arxiv.org/pdf/2304.03271. (AI water-use estimates and the frequently misquoted ‘bottle of water’ claim: training GPT-3 in Microsoft’s US data centres estimated to have consumed millions of litres of water; roughly 10 to 50 medium-length responses could consume about a 500 ml bottle of water, depending on when and where the model is deployed; water efficiency varies spatially and temporally.)