Assessing the Climate Impact of Various AI Prompts

A significant portion of Americans regularly interacts with artificial intelligence, but many are unaware of its environmental costs. For instance, using ChatGPT consumes ten times more electricity than a Google search. In 2023, data centers powered AI models accounted for 4.4% of U.S. electricity use, projected to rise to 6.7 to 12% by 2028. A recent study in Frontiers analyzed various large language models (LLMs) and found that more complex prompts can emit up to 50 times more CO2 than simpler ones. The study noted an accuracy-sustainability trade-off, with the most accurate models producing disproportionate emissions. It highlighted the disparities in emission rates among different models; for example, DeepSeek R1’s responses equated to the emissions of a round-trip flight from London to New York, while Qwen 2.5 could answer three times as many queries with similar emissions. Researchers urge users to consider the environmental impact of their AI usage.

Source link