Dan Romero on Warpcast

Dan Romero pfp

Virtually all of the people on Farcaster with large audiences: 1. Don't have large audiences elsewhere (Farcaster is their first time with a large audience) 2. Didn't know me or Varun and didn't know each other. The "Farcaster OG scene" is net new creation of Farcaster. 3. Were early and stuck around when everyone made fun of Farcaster (and people still do) 4. Continued to cast interesting stuff, casted in channels when we suggested people cast in channels 5. Warpcast onboarding asks all users for their interests 6. Based on the interests, we pair the account with an initial set of follows 7. There are 200+ accounts on the list, weighted toward accounts that get the most engagement / drive retention, but a large group of people benefit 8. On our list to improve, but the reality is outside of "I should be on the list!" haven't seen any concrete proposals to improve

48 replies

18 recasts

167 reactions

androidsixteen pfp

@androidsixteen.eth

On #8, it's hard to propose a concrete proposal when the auto-follow list and process for determining it are opaque If there was more info on how it worked, I think people might be able to suggest modifications. Otherwise, you either get polarization for or against the black box

1 reply

0 recast

16 reactions

Dan Romero pfp

1. Look at the list of the top 100 most followed accounts on the protocol. 2. Suggest other accounts that should be on that list. 3. Obviously that will have a strong personal bias, but if no one is willing to take the time to do that, my sense is the list is good enough (sure there can be 1-2 improvements, but that's less of a priority vs. 50/100 being wrong).

4 replies

0 recast

7 reactions

androidsixteen pfp

@androidsixteen.eth

Is it just a random selection of 50 from the most followed 100 (or 200) at any given time?

1 reply

0 recast

0 reaction

Dan Romero pfp

It's based on what interests you select and the overall engagement each person gets, historical output / quality (subjective pov, but so was the onboarding approach for the first 10K). Then there is an additional set of users that is random in both number and users that get added at the end.

1 reply

0 recast

7 reactions

Sid pfp

I think the problem is more in weighting + concentration rather than choosing a specific 100 per se. The dispersion of the auto follow list can at the minumum be spread among the broader range of the ~3K or so active casters (PBs) rather than just 60-100 (accounting for user interest selection at onboarding) Some concentration stats: - The top 20 users have on avg 50% of the entire Farcaster userbase following them (~300k out of 572K) - The top 50 users have on avg 34% of the entire userbase following them (avg 200K followers)

1 reply

0 recast

0 reaction

Dan Romero pfp

What if that increased dispersion lowers d30 retention by 5%? What's the optimal choice for the network?

2 replies

0 recast

0 reaction

Sid pfp

1. Not sure if you'e already trialled but worth running an experiment here to see if material change to d7/d30 retention rates? Goal here being to encourage broader range of long tail casters who cast in subject specific channels to continue posting interesting content - potentially flattening curve from d7 -> d30 retention. 2. Optimal choice for the network would be to experiment to verify what optimizes for growth + flattening of retention curves. Broader range of incentivized casters = more word of mouth growth + more potentially interesting content across subjects.

2 replies

0 recast

1 reaction

llamafacts pfp

@llamafacts.eth

What if it doesn't? Have you tried? I think the only way to know is to try this and see new user's behavior based on different algorithms. Maybe you'll find a way to increase retention/engagement or motivate other users to cast more often if they get new followers.

0 reply

0 recast

0 reaction