Oddbean new post about | logout
 It’s really basic, just a fixed list. We would love to improve it if anyone wanted to take a stab at something better. 
 I could create an ideal formula but coding does not come from these fingers unless its stealing bits of htlm 
 I know Will is not the biggest fan, but could be people discovery DVM 

https://www.data-vending-machines.org/kinds/5301/ 
 Don't DVM algos just serve up a variation of the fixed lists, since they look at what people are already looking at, and everyone is looking at the fixed list?

I don't think we're going to get around actually analyzing the content, rather than the content-providers. 
 It really depends. They could dynamically analyse some recent (popular) content, group it into categories and suggest users that wrote that content. This way suggested users would be up to date and not some chosen people forever, regarding if they are active etc. 
 Yeah, but that's the Catch-22 I mean. You have to already be popular, to have popular content. That's why follow lists have been getting more and more similar, with recommended or trending content often being the same 5 people over and over, with every post they make.

Gigantic bifurcation in the number of followers (200k versus 20), is the cause, and we can't end the cycle until we figure out how to surface unpopular-but-interesting content. 
 I mean, "popular" can be defined in many ways.   It's really a matter of implementation. It can be absolute, but it could also be adjusted in terms of time/follower number etc. it's really a matter of implementation and what is wanted.