Resolves using a poll ran in December 2024, to the most impactful arXiv poster first posted in February (revised versions don't count.)
A possibly related search in Semantic Scholar (note this goes back to 2020, as the results aren't terribly good if the date range is narrowed further): https://www.semanticscholar.org/search?year%5B0%5D=2020&year%5B1%5D=2024&venue%5B0%5D=arXiv.org&q=%22machine%20learning%22%20OR%20%22artificial%20intelligence%22%20OR%20%22AI%22%20OR%20%22neural%20network%22%20OR%20%22neural%20networks%22%20OR%20%22language%20model%22%20OR%20%22language%20models%22&sort=influence
@duck_master I'm trying to see if manifold can help surface good papers out of the literal hundreds a day being posted to arXiv. Not sure semantic scholar has the ability to make judgement calls on such a short time scale since it relies mostly on citations.