Training a single AI model can emit as much carbon as five cars in their lifetimes
Source: MIT Technology Review
The University of Massachusetts, Amherst conducted a study revealing the substantial environmental impact of training large AI models for natural language processing. The research found that the carbon emissions from training these models can exceed 626,000 pounds of carbon dioxide equivalent, significantly surpassing the lifetime emissions of an average car. The study focused on four prominent AI models: Transformer, ELMo, BERT, and GPT-2, highlighting the escalation of energy consumption and environmental costs associated with larger model sizes and additional tuning steps. The findings shed light on the urgent need for the AI industry to address the environmental consequences of deep-learning processes to mitigate its carbon footprint.
More detail: https://listentoyourmuse.com/ai-carbon-footprint/power-hungry-processing-watts-driving-the-cost-of-ai-deployment