The holy grail would be generating TL;DRs with a model trained on the best human-written abstracts. Although - there should be a way to hide machine-generated content, for your convenience, since not everyone might be interested in this content
No. The holy grail would be TL;DRs written by the actual author, at the time of writing the actual article. No one can better know what the important points are and how to best highlight them. The work of producing a good abstract is a tiny fraction of tiny fraction of the work involved in producing a good article.
Community-written abstracts also have their merits too. The author's perspective might be limited to their area of expertise and is therefore biased. That perspective can be normalized by the more nuanced interpretations from the community.
I feel like this is basically already what people who promote their papers on twitter are doing. Other than Twitter or LinkedIn or something (I don’t really use either) I’m not sure there’s really a place for researchers to actually post such a summary where it would be useful, unless everyone starts using this site I guess.
Even then, I feel like my time as a researcher would better be spent writing a better abstract, so that this sort of tldr could just be a bolded first or last sentence of a conventional abstract… but I’d be very interested in what other people see as the important or most interesting aspect of my paper
Along these lines, I’ve recently taken to using revtex structured abstracts, which are basically a series of topical tldrs covering context, purpose, methods, results, and conclusions. Maybe I’ll experiment with adding a global tldr at the top
