I recall X dogpiling on this ValsAI result when it came out, with the community note and all (which I rated as unhelpful for "missing the point").

It's actually cool as hell. And lean verified too. Dijkstra is provably optimal *except* between m = O(n/log n) and m = O(n log n), so any progress in that range is a breakthrough.

I'm going to turn it into a bigospeedrun race.

Quoted post by Vals AI: x.com/ValsAI/status/2102470503328010349

Quoted post by Nil: x.com/Nil053/status/2103876483483938962