I'd really like to know more about this -- the replies to the original question were divided between people commenting about how much better the comments were, and people talking about better tooling and empowerment being granted to large YT creators.

My instinct is that the progress has been primarily because of the better tooling and moderation capabilities, just because the larger companies always promise that ML will take care of their large-scale moderation processes without much concrete evidence. But I also suspect that the answer is somewhere between the two approaches, and it would be great if YT were able to improve the wider ecosystem by sharing some of their strategies.