More evidence against algorithmic rabbit holes/echo chambers: relying exclusively on the YouTube recommender results in less partisan consumption, not more. Evidence from a very clever new study comparing real-world consumption patterns with "counterfactual bots"
www.pnas.org/doi/10.1073/...
This is a tunable parameter in recommender systems, showing similar versus surprising, focus on familiarity versus discovery. This result argues YouTube's algorithms as currently tuned are good, but says little about their algorithms at other periods of time or if others have tuned theirs poorly.