@jfpuget But actually even if I get that data, what's the conversion? Is one negative use worth 10 positive ones? 100? 1?

So even with data, it is still a matter of opinion :)

@jfpuget I don't think there is a way to prevent something bad to be said wherever something can be said.

So the only thing this is about is whether this feature would be used more positively or negatively, compared to other features. And I don't think there's any data on this, so at the end of the day, it's about everyone's (or rather, the dev's) anecdotes and personal preferences.

Though it could be a fun side project to gather this data from twitter at scale!

@ducha_aiki @at oh lol they show this sketch, but we actually have a real figure like this in our paper!

We have some updated version of this figure with newer models for talks and it did bend down slightly.

I think the authors probably don't know or paper but would enjoy it! I should email them, but maybe @rami or @erturk (only two authors I found here) can share this with their co-authors and save me the typing of many emails?

And agreed with you re small data.

@ducha_aiki that's 2yo, I'd definitely do separate arxiv.

A two-pager is not too rare/weird, off the top of my head: "fixres: efficientnet", "mocov2", and our own "better plain vit baselines" are all like this

@jfpuget I still disagree with it. Screenshot below is the argument for not supporting them.

I want to use them for exactly the mentioned reason, but in a positive way! I want to add my comment to the quoted post while sharing it with my followers and giving proper credit to the author. This is brother achieved by screenshot not by replies, but it's common academic discourse.

Sounds like the author has some bad experience with trolls and only sees the potential for abuse, that's sad :-/