Cover photo

Moderating a Farcaster channel

What I have learned from moderating the AI art channel

This is my third and final article in a series about Farcaster channel moderation. The first article described how the new decentralized moderation works and why it is different from what is available on other social networks. The second article explained the mechanics of how to set up moderation with automod. And this final article will provide some thoughts about how to moderate a channel. I am writing this from the perspective I've gained moderating the /ai-art channel, but I think the concepts are general and will carry over to many other channels.

As a way to organize our thoughts around moderation, it may be helpful to categorize users into a few commonly seen categories. The following are five types of users I routinely recognize:

Blatant spammer. These accounts just spew random stuff into your channel.

Engagement farmer. These accounts pretend to post on-topic, quality content while actually just providing low-effort stuff intended to maximize engagement.

Quality contributor. These are the accounts that get everything just right.

Oversharer. These accounts could be quality contributors in principle, except they post too much.

Newcomer/guest. These accounts may be contributing interesting material but they are too new or cast too infrequently for the moderators to know and recognize them.

The main job of moderation is to elevate the quality contributors and the interesting contributions of newcomers or guests while rate-limiting the oversharers and not being tricked into elevating the engagement farmers. In a moderation system focused on banning/hiding, you would try to identify the blatant spammers and engagement farmers, ban those, hide some portion of the posts from the oversharers, and let everybody else post at will. Farcaster is doing the opposite currently and emphasizing curation rather than hiding. In this context, you will want to elevate the casts of quality contributors, as well as the interesting contributions from newcomers and some fraction of the material from oversharers.

The blatant spammers are easy to identify. Their casts are obviously inappropriate or off-topic, and their cast volume tends to be high. The engagement farmers on the other hand are more subtle. On first glance their casts may seem reasonable. I've found this to be particularly problematic in /ai-art, where people just spin up ChatGPT to generate a constant stream of motivational quotes with accompanying images. The resulting casts can look very convincing but once you've seen a couple hundred of them you recognize them right away.

The quality contributors are typically the reason you visit a given channel. Their casts tend to be on topic and interesting, and they cast somewhat regularly yet not so much that they dominate the channel. The oversharers don't look that different from quality contributors except they post too much. If you don't slow them down, they'll just take over the channel and drown out all other material. Finally, the newcomers or guests tend to not be known to the moderators, so their casts need to be evaluated carefully on a one-off basis.

What are best practices to implement moderation? In principle, a single person could moderate a channel, but in practice I think it's too big a job for one person, even for a small channel. You need a team. And even then, people get distracted and have other things on their plate and forget to moderate so you may want to automate some moderation. One thing you could consider is allowlisting. Maybe all moderators could be allowlisted. If you can entrust them with moderating Main you can probably also entrust them with casting into Main. Next, you could consider identifying specific quality contributors and allowlisting them. There's a tradeoff here between the size of the allowlist and the amount and type of work the moderators have to do. The smaller the allowlist, the more moderators have to find individual casts in Recent and curate them, and the larger the allowlist, the more moderators have to monitor Main and potentially hide casts from there. Which approach is more appropriate may depend on the channel size, the moderation team, the composition of the channel audience (small, close-nit group of frequent contributors versus large community of occasional contributors) and also simply personal preference of the channel owner.

I also think allowlisting could be used more aggressively if we could apply a rate limit to allowlisted accounts. For example, each account might be allowed to cast two casts per 24 hours into Main but any further casts in that period would be hidden by default (but could be rescued manually by moderators). This would be a great way of moderating oversharers, as they could be allowlisted and at the same time rate limited. This is not currently possible but the Farcaster moderation tools are under heavy development and I hope this type of moderation will become available eventually.

I'd like to point out that the default Warpcast moderation setup, where casts from accounts with power badge are always shown in Main, approximately implements the strategy I have outlined here, by making the assumption that all holders of a power badge are quality contributors and should be allowlisted. And I'm afraid this will frequently not be a good assumption. Even if somebody is a quality contributor in aggregate, taken over their entire activity on Farcaster, they may be an oversharer, engagement farmer, or simply a confused guest in one specific channel. So I feel confident allowlists will generally have to be constructed specifically for individual channels, or maybe for families of channels. (Various different art channels could share allowlists, for example.) We definitely need better tools to construct such allowlists if we don't want to do this entirely by hand.

Let me end with some thoughts about moderating /ai-art specifically. Since it is so easy to generate AI images, this type of a channel can get overrun with tons of low-quality or low-effort material. Now of course the moment I say this, people will complain that effort is irrelevant for art. Somebody can write a three-word prompt that produces outstanding art while others may put tons of effort into their work but the outcome is mediocre or irrelevant. I agree. That's why I tend to evaluate effort at the level of the account rather than the individual cast. I try to determine whether the individual has a consistent artistic practice or instead is just engagement farming on Farcaster. I won't specify in detail how exactly I do this, but suffice to say I look for consistency, a presence outside of Farcaster, engagement with other artists I know, discussion of methods, techniques, or concepts expressed in the art, and so on. Accounts that lack all of this and instead just constantly post their latest AI-image mint on Zora are unlikely to score high on my "consistent artistic practice" scale.

Of course any one assessment I make may be wrong or biased. That's why we have a moderation team. I may have a bias in favor or against certain accounts and others on the team may have the opposing view. That's great. I don't think moderation has to be particularly strict to be effective. In case of doubt, it's better to let more content and diversity into Main. There's still the added level of moderation coming from Farcaster's algorithmic sorting. It gives all the channel subscribers the power to increase or decrease the visibility of certain casts, by liking and interacting with them on the one hand or ignoring them on the other. So, as you're browsing the feed, please don't think that because there is now manual curation you no longer have to interact with casts. Please continue to like and recast the best casts you come across. It will help their authors and it will improve the overall feed quality on the platform.

Collect this post to permanently own it.
Claus Wilke logo
Subscribe to Claus Wilke and never miss a post.
  • Loading comments...