Language models are notorious for their high energy consumption. Must AI applications inevitably become ever larger, and will this actually lead to improvements? This presentation discusses this question and explores the aspect of an AI paradigm that prioritizes efficiency.