AI-powered
podcast player
Listen to all your favourite podcasts with AI-powered features
Is There a Trade Off Between Time and Efficiency?
There is a trade off between how much time you put in as an engineer, or as a researcher, as any kind of practitioner. You could really carefully narrow down your hyper parameter ranges and then spend less in gpu hours to find some good optimum. Or you could just set it up to be a superbroad search, let it run for a week, and it'll take you two days less time to run those experiments of your own hours. The key idea, i think, behind our green ai paper is that this happens all the time with people, and often we just don't report that. We'd like to encourage organizations to continue to release their models in order to save others