Innovations For Energy Efficient Generative AI

While the explosion in the capabilities of generative artificial intelligence (AI) and the associated benefits has received a tremendous amount of attention, both in the technology world and popular press, so has the unprecedented amount of compute and energy required to train and serve these extremely complex models. Many of these concerns are well founded given the hundreds of megawatthours required to train large language models (LLMs). However, our contention is that history and recent developments indicate that our industry’s ability to innovate is up to the task of delivering the benefits of LLMs while reining in energy use. This paper will review some of that history and the approaches that are likely to provide the required efficiency gains.