The study also tested several AI image generation models, including Stability.AI’s Stable Diffusion XL, which was one of the least energy-efficient. It did not include the most popular models like DALL-E or Midjourney, but noted that these larger, more widely used models are likely to have a greater environmental impact. Dr. Sasha Luccioni, who led the study, emphasized the need to be conscious of where and how we use generative AI, comparing its cost and benefits, and suggested that smaller models could be used in areas like web search and navigation due to their large energy requirements.
Key takeaways:
- A new study by Hugging Face and Carnegie Mellon has found that creating images with generative AI models can use as much energy as charging a smartphone, and popular models like ChatGPT’s Dall-E and Midjourney may produce more carbon than driving 4 miles.
- Dr. Sasha Luccioni, who led the study, emphasizes that every time we query an AI model, it comes with a cost to the planet, and it’s important to calculate that.
- The study found that large, multipurpose models, like ChatGPT, are more energy-intensive than task-specific models, and image generation took substantially more energy than any other task for generative AI models.
- Dr. Luccioni suggests that while multipurpose generative models are user-friendly and easier for consumers to work with, they are more energy-intensive and their deployment should be limited to contexts where tasks are not well-defined.