ChatGPT, which entered our lives in November and continues to remain in our lives both on its own and by being added to many services, does not seem to be very successful in resource consumption. According to a study, ChatGPT raises questions especially about water consumption.ChatGPT’s thirst to end
ChatGPT and similar LLMs use up to 500 milliliters of water for every 20 to 50 prompts or questions that users ask, according to researcher Shaolei Ren of the University of California. In an article published on Arxiv, the researcher noted that 500ml may not seem like a lot, but the water usage is enormous with ChatGPT being used together by users globally. The number of unique people who visited ChatGPT in August was 180.5 million. The total number of visitors is 1.43 billion. On the other hand, it was reported in the previous months that the amount of water used in the training of the GPT-3 model was 700,000 liters.
The verification of this research is actually provided by Microsoft (Microsoft is the biggest investor of ChatGPT creator OpenAI.) It is stated that most of the water increase in Microsoft’s 2022 environmental report is due to artificial intelligence. Microsoft’s water consumption increased by 34 percent to 6.4 billion liters in 2022. Microsoft, on the other hand, said it has acknowledged this issue and is exploring ways to measure energy use and carbon footprint. Internally, the software giant is also working to make major language models (LLM) less energy-intensive, such as ChatGPT (GPT-4, 3.5, etc.).
Microsoft plans to achieve carbon negative, water positive and zero waste targets by 2030. However, OpenAI also acknowledged the water use issue and said it was looking for ways to make LLMs more energy efficient. While the amount of water used to power LLMs is pretty staggering, it’s worth noting that these modern LLMs are not that old, so over time methods can be developed to reduce water and energy use.
59076
