4 ms·

(tangent) I wonder if there's a few smart reasons CS papers aren't "required" (or at least culturally expected) to put their working code on GitHub (with appro

• PaulHoule 4y ago
  You get ahead in academic CS by writing papers not by writing programs. Some people in CS are brilliant programmers but I’m certain that a good fraction of CS papers didn’t really implement the algorithm that they thought they implemented. (No different from the pervasive bugs in application software.) If the paper is the product than the considerable work involved in documenting something and making it reproducible (easily π times the work of a spike implementation). Sometimes you don’t need fully working code, if you can prove that something you think might work really won’t work that is a very useful research paper but it obviously does not end in working source code. I try to post papers from arXiv to hacker news with an eye and quality and thinking about what makes a good submission. One of them is length. A lot of CS papers are just 4 or 5 pages and I think almost always papers of that size don’t have enough room to really explain anything, particularly to a lay audience. I guess I could write a script that downloads the PDF and counts the pages but back when I worked at arXiv years ago they had rate limiting that would make it hard to write a script that would do that and I suspect they still do. I can imagine downloading the PDF for 20% of the articles but not much more than that. A link to working source code would be another good quality indicator.