PI Probaligence Gmbh

"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."

The MIT Technology Review published an essay about the carbon footprint of AI image generation, emphasizing the wast resources especially in the usage of large models for image generation. 

We at PI believe the future of AI lies highly specialized, efficient and problem specific solutions and not – as the current trend seems to be –  in just bigger ANNs for general purpose.

The carbon footprint of large, general-purpose ANNs is a pressing concern. Training these models involves extensive computations, often requiring vast amounts of data and powerful hardware infrastructure. The energy demands of such processes contribute significantly to the carbon emissions associated with the computing industry. In contrast, small, specialized ANNs, tailored for specific tasks, demand fewer resources and can be trained with a smaller dataset. This results in a reduced carbon footprint, aligning with the global imperative to adopt sustainable technologies.

Furthermore, the deployment of small, specialized ANNs can lead to more targeted and effective solutions. By focusing on a narrow range of tasks, these networks can achieve higher accuracy and performance in specific domains. This not only enhances the quality of outcomes but also minimizes the unnecessary computational overhead associated with larger models.

Find the MIT Technology Review article below:


Stay Updated

Join Our Newsletter

Get regular updates about PI’s new developments and activities in the field of AI and machine learning.


Social Networks

PI Probaligence GmbH

(c) 2023 All Rights Reserved