However, it also seems to interpret "hate viewing" as "high interest" in terms of engagement. Sometimes I'll look at an account just one time, often by searching for it, in order to see some Tweet that's in the news or generating controversy. And then a few days later I'll start seeing Tweets from similar accounts in the "for you" tab, which I definitely don't want to see.
On other websites with recommendation algorithms, like YouTube, I can avoid this effect by viewing ragebait content in an incognito window, effectively curating my algorithm by opting out of it for content I don't want affecting future recommendations. But I can't do that with Twitter, because the login wall requires authentication to view more than the top three replies to a tweet. I wish there were some way to give the algorithm more intentional feedback on its recommendations.