Ask HN: How to build better, more useful recommendation algorithms?

I just saw this video (https://www.youtube.com/watch?v=jUCx6-WNDnA) and recommendation algorithms have always interested me. They are a huge factor in the way we perceive information on the Internet and also affect our opinions in general.

There is some sort of positive feedback loop where algorithms recommend us the exact same content we have already liked instead of showing us other new subjects or viewpoints.

What are some alternatives to this polarizing recommendation model and what do you think of them?