[nseggert]

This is definitely great foundational work, but if you click through to http://ann-benchmarks.com/, you can see that they show dominant performance on the glove-100 benchmark, but decidedly middling performance on the other benchmarks for which they provide results (glove-25 and lastfm-64). I don't see this doing anything to unseat HNSW as my go-to algorithm for its strong performance across a wide variety of benchmarks.