The MIT Technology Review published an essay about the carbon footprint of AI image generation, emphasizing the vast resources consumed when large models do the work.
The team found that using large generative models to create outputs was far more energy intensive than using smaller AI models tailored for specific tasks.
MIT Technology Review
We at PI believe the future of AI lies in highly specialized, efficient, and problem-specific solutions, not, as the current trend seems to be, in ever bigger neural networks for general purposes.
Two reasons smaller wins
- Energy. Training a large, general-purpose network takes extensive computation, vast data, and powerful hardware, and those demands are a significant share of the computing industry's carbon emissions. A network built for one task trains on a smaller dataset and demands fewer resources.
- Results. Narrowing the range of tasks raises accuracy and performance in that domain, and avoids the computational overhead a larger model carries whether you need it or not.
Read the essay: Making an image with generative AI uses as much energy as charging your phone, MIT Technology Review.
