At least in my personal observation, it seems like star ratings elsewhere on the internet (I'm thinking particularly of Airbnb and Google Maps reviews) are probably pretty close to bimodal: by and large, people essentially use five stars as a "like" button and one star as a "dislike." Assuming this is the case, a rating of 4.8 stars is essentially a way of saying "95% liked this": why not just strip out the extra abstraction and complexity and instead use a simpler like/dislike model, just like that YT blog post argues?

Of course, there are other situations where ratings between 1 and 5 stars _are_ informative: think critic reviews of movies, video, etc. But I think most crowdsourced reviews have kind of converged organically to a like/dislike model.