Examples of this might include prioritizing mutual followers on Mastodon, or prioritizing low-traffic subscribed communities on Lemmy so that they don’t get lost in the 50 posts from the busier communities.

Also:

3. A open, customizable algorithm that lets the user set their own priorities, and if it does any “learning” based on user actions, it’s geared toward the user’s priorities and easy for the user to see and correct what it’s learned.

Again, key factors being: open, customizable, correctable, and serving the user, not serving the platform.