2014-15 was when the news feed switched from static, human-written code to machine learning models optimized to maximize ad revenue[1], the models discovering "user attention time" was a far better metric to optimize for than clicks, likes or shares[2]. And what people spent the most "attention time" on was content that triggered raw moral outrage, fear and tribal hostility[3].
This switch does appear to have put off a lot of smart, thinking people as the newsfeed degenerated into a torrent of dopamine-spiking diarrhea. But those people don't tend to be the ones that click on ads.