Having false positives are quite serious though, especially when it comes to shutting accounts down at scale. Sure it works on their small sample test runs, but anyone who's applied these algorithms in real life in an automated fashion understand the risks of false positives is usually far higher in practice.

You will basically be put at risk for following 'controversial' accounts on Twitter of having your account delete accidentally. And businesses will have to jump through bureaucratic hoops to use legitimate automated scripts to manage accounts.

It also seriously deters anonymous online speech if they force a big percentage of users to provide a unique phone number to prove they aren't a bot (the current Twitter strategy for countering spam). Not to mention the liberal sharing of information with nation states, surreptitiously or otherwise, even for people not trying to be anonymous. A phone number is a critical identifier in the surveillance industry.

Getting this accurate is very important and I feel like that is being heavily downplayed in this article.