Were you measuring what users like, or what they "engage" with?

If I see a post on mastodon, it's from someone I follow or a post that someone I follow thought was worth sharing. That's a genuine interaction between humans. I get to decide what's in my feed based on who I follow and the filters I set up. The system is designed only to connect me with people I choose to follow.

Meanwhile, an algorithmic feed with no options is designed for one thing: manipulating users to optimize "engagement". That by definition requires ignoring what users prefer. I really shouldn't have to explain that this is a bad thing.

We ran lots of experiments, including asking people survey questions, monitoring usage, asking people offline to try both, etc

You're wrong about the design of algorithmic feeds. I worked on them, that's not how we designed them.

How about giving the people the option and letting them decide for themselves.

Also, 20% of people is hundreds of millions of people.

Facebook engr is famous for popularizing the phrase “one in a million happens every millisecond” so this 20% hand wringing is funny to me

People have the option, you can choose to use a chronological feed on Facebook and Meta

I mean, yeah, but you could also do a lot of high powered randomized trials about kinds of cookies and learn that 80% of people always prefer eating the cookies with cocaine in them.

That's a different argument, and not one I really care about. It's no more "addictive" than tv and way less harmful

To push back, it does seem like you care. You've created rationalizations and offer them up unprompted. I think most people would argue for allowing both settings and having the app default to the one the internal research found more preferable and more profitable. Not offering the setting is telling.

I meant I don't care to argue. I've had this conversation too many times, and it does have the setting. You can Google it to figure out how to change it.

I think the point is: were the experiments testing whether people said they liked X better and immediately engaged or were you testing whether or not the user felt good about the experience and wanted to come back long term.

My experience is meta prioritizes instant gratification, which yeah people and their immediate actions says they want. But also I’ve completely stopped using Facebook and instagram because it became clear it’s of limited value to me. Yeah, i might mouse over a piece of salacious content because it’s salacious. But i know it’s low value to me. And most of my peers are in the same boat.

We ran experiments that lasted up to 5 years. People who have ranked feeds self report a better experience after using the product for years, although you have to be careful with these sorts of things because the chronological users are more likely to churn, which biases the results

Interesting. I just logged into facebook and my feed is spam, and my notifications are basically a chronological feed of my closest connections.

If you don't use Facebook regularly, the feed will be bad. They use almost no data older than 90 days

20% is far from small...

Typically less, depends on the surface.