News for hackers. Stuff that matters.

cm228

depends on the chunk-size used to create the embedding.
they cluster the examples with their model and then check the predictions against the labels.