Environmental Impact
Training and running state-of-the-art generative AI models consumes large amounts of electricity and water and produces carbon emissions, concentrated in the data centers behind the biggest models. This is a harm that often falls hardest on already-burdened communities, and it is a major reason small, local models are appealing: they can run on ordinary hardware at a fraction of the cost. Asking whether a task genuinely needs a large model at all is partly an environmental question.