This site is the bread and butter of each Research Engineers and Scientists working in Deep Learning. You use the site almost everyday.
Advanced learners also use the site regularly.
You would just think that "everyone knows" and never think of sharing the site on HN.
Knowledge about a field transfers best by hands-on association with people who practice it. Before widespread IT, communities of practice were local and relatively homogeneous; so it was easy to share the essentials of a field quickly, and get newcomers up and running with best practices.
Nowadays however, communities of practice are widespread, coming around all the world with very different backgrounds, communicating through low-bandwidth channels, and we're flooded with information so it's difficult to ascertain what is essential and what's accessory.
It is much more difficult for an outsider to grasp the essential qualities of a field they want to enter, as there are usually no guides comprehensive enough to detail everything you need to know.
I never found any subject that needed let's say more than 10 minutes of internet searches to know if it's worth pursuing.
It was much harder before the web. I remember as a kid seeing books about C++ in the local shop but even with looking inside not understanding what C++ was. Nowadays I would get my answer almost instantly.
You couldn't possibly believe this if you were old enough to remember what preceded the internet.
Good lord, no, today is not worse than microfiche and card catalogs.
It was still possible to define a Library Science were books were classified by hand, and not some secret algorithm counting links as votes or learning and regurgitating a corpus of loosely related documents without understanding any of it. I.e., it was possible to make sense of information sources, and whatever you learned of a field came with a single consistent narrative. Nowadays, information gathering has become an exercise in picking and choosing unconnected fragments from which you must infer your own understanding.
In some ways you can still try to emulate the old way, by limiting yourself to a small set of publishers who try to compile and organize a small part of a field of knowledge - yet it is much easier than the teachings of that source will be deeply contradicted by some other seemingly authoritative source, without a clear way to know which one should you rely upon, and with the whole exercise feeling like it provides an incomplete perspective.
“Bread and butter” for me is http://arxiv-sanity.com
https://news.ycombinator.com/item?id=19054501 (Feb 1, 2019) 411 points, 23 comments
https://news.ycombinator.com/item?id=23391934 (June 2, 2020) 304 points, 21 comments
Pre-parenthesis part is dead serious, parenthesis part is slightly hyperbolic due to accumulated trauma with bad reviewers
Another great resource is the HuggingFace model zoo. So many trained models easy to deploy.
- Checking the state of the art (SotA) for a given problem. For some problems 2 year old solutions are still close to SotA; in others - there is a huge difference. And if there is a huge difference - is it because of architecture and parameter tuning, or using totally different architectures and training modes.
- Running code - to be used somewhere, or as a reference. Papers never have all details, and do not compile.
Context: I used to work in the field, as a consultant. Though, I cite Papers with Code in one overview paper.
"Ten Thousand"