Technical discussions of new research seem to have mostly disappeared in this subreddit, because researchers became a small fraction of its immense readership of 3e6 members.
So I created a subreddit to host such discussions. A “safe space” for researchers, if you will, with strict standards for content^1 . I seeded it with posts about a few recent papers I thought were interesting and my own takes on them, to get the discussion started.
But then I said to myself: “You don’t have time to manage a subreddit. WTF are you doing?” and deleted it all. Nevertheless, I’d like to see someone else, perhaps someone with more time, try to do it.
^1: Its main rule was: “No low-effort or low-expertise posts or comments: If your average ML PhD student, or someone with a higher level of expertise wouldn’t have posted something, then it does not belong here.” Other rules dealt with the format of the posts.
We have technical discussions of new research here constantly.
These are just my posts and there are many others:
https://www.reddit.com/r/MachineLearning/comments/17ofere/r_diffusion_might_be_a_better_way_to_model/
https://www.reddit.com/r/MachineLearning/comments/17geq9f/r_researchers_discover_that_incontext_learning/
https://www.reddit.com/r/MachineLearning/comments/17aycuw/r_meta_ai_towards_a_realtime_decoding_of_images/
https://www.reddit.com/r/MachineLearning/comments/175ep6x/r_tsinghua_university_inverting_transformers/
Fortnightly. Finally got a chance to use this word :-) 4 links spanning 2 months.
But even in these picks, take a look at the first one, for example. 10 comments. Only one of them suggests that the commentator looked at the paper itself.
Those are just **four** links **I myself** have posted and they have ~75 comments on them. I pointed that out in the original comment but you skipped over that.
But the point you were trying to prove was that the discussions were “constant”. How does picking your own threads spanning 2 months support it at all?
The OP didn’t say that the discussions were completely gone. Yes, there are some, but pretty thin and usually glib. I don’t count “Wow! This is exciting. I’ll have to take a look at this awesome new paper!” as discussion. A bot harvesting upvotes could post this.