"Getting shit to work" is more foundational to machine learning than you would think, and containers helps a lot with that. If you want to train models on someo

"Getting shit to work" is more foundational to machine learning than you would think, and containers helps a lot with that. If you want to train models on someone else's machine - and you probably will, for anything big - you need to know a little about how that sort of thing is done today. And if you want to try two different deep learning frameworks, dependent on different versions of cuda, and want them to not break each other, God help you if you try that without containers. It's not that they don't have a "course in docker". I understand that. It's that they haven't even heard of it, so they don't even know where to start to look for solutions to problems like that. I have been through that pain myself. Containers is just one of so many easy things, that make your job so much easier, I've learned the painful way in 20 years as a developer in (mostly) non-elite companies, where no one else knew it either because they hadn't been taught at the local universities, because no one there knew it either.