Communities · Study simulations · Graph

Clusters

loading…
The verdict
What's wrong with today's clusters
The new way to build them
Would you have studied the right facts?
Per category, at a budget of
Your same-fact ratings
Where "same fact" starts
Share of the pairs you rated the same fact, by how similar the two clues' meanings are (embedding cosine), with and without a shared distinctive phrase.
Pairs you judged where the two methods disagree
Per category
The five categories with a full run of the new method. Browse any of them side by side.
What's next
How this was measured