Moderation team, how we work and our rules choices

Primfeed's Moderation Team consists of seven individuals dedicated to ensuring that all content uploaded to Primfeed adheres to our community guidelines and Terms of Service.

Written By Luke Rowley

Last updated About 10 hours ago

By following these standards, we aim to create a positive, respectful environment and social space for Second Life™ residents.

Our team reviews every piece of content manually. This human-focused approach allows us to provide thoughtful assessments of each content uploaded, ensuring that all content adheres to our Terms of Service.

The moderation team's goal is not to police your content but to ensure consistency and safety across the platform.

How do we work?

Primfeed’s moderation team has a dedicated Discord server with different channels:

  • #mod - our main channel to talk about various things regarding Primfeed and the moderation.

  • #todo - Luke’s todo. Moderators don’t have full permissions on Primfeed and various actions are still limited to Luke.

  • #rollback-content - we use this channel to post content we roll back after a moderation took an action on a post. This ensures every moderator knows what was rolled back and why.

  • #is-this-ai - we use this channel to ask if a post is AI or not, to have a second opinion when we’re not sure.

  • #reports - this channel is automatic, and we receive all reports users submit here. The moderator adds a ✅ or ❌ to let the others know whether they've checked the report.

  • #moderation-logs - this channel is automatic, it shows all actions taken by moderators.

What are our biggest challenges?

Amount of content to moderate

In two years of existence, the number of pictures uploaded to Primfeed every day has increased by ~60%. While we're very grateful for the platform's success, it comes with the challenge of keeping up with the volume and staying efficient enough to moderate everything.

Uploaded media per day, since July 2024

To stay efficient, we rely heavily on the report system built into Primfeed, whether it's to report a post or a profile. Moderators aren't full-time employees and they don't sit at a regular 9-to-5 moderating Primfeed. They work flexibly rather than fixed hours, because forcing a set schedule wouldn't be fair to them.

Why don’t you hire more moderators?

That’s a great question and the simple answer is trust. The current moderation team has proven to be completely trustworthy. They don’t want to create drama nor they are looking to abuse their “powers”.

For now, we don’t feel there’s any need to hire more moderators. If Primfeed keeps growing and the moderators themselves say that they can’t keep up anymore, we will open the recruitment for more.

Detecting generative AI

AI generated or altered content is our biggest challenge. Primfeed has two rules about AI content:

  • If it’s commercial and the product sold has been altered or generated by AI, then the post needs the AI label.

  • If it’s not commercial, AI usage is not allowed in the Public Gallery.

    • Exception: if the background used isn't obviously AI-generated, we may accept it as long as your avatar takes up a good portion of the picture. This exception exists because more and more bloggers buy backgrounds from the Marketplace or online and don't always recognize AI-generated content.

We rely exclusively on our human eyes to detect generative AI because AI detection tools aren’t reliable. Our team members have different levels of experience with detecting AI but overall, we have 1-4 years of experience with AI generated content and the “youngest” moderator has 15 years in Second Life™.

Generative AI gets better every single day, and that makes detection harder over time, not easier. What works in our favor is Second Life™ itself. Second Life™ is a game and not a realistic one, even though it keeps moving in that direction. It has its glitches, its small rendering errors, and a specific kind of texture, while AI tends to generate real-life content, and that's one of the biggest giveaways. Generative AI also has its own glitches and recognizable patterns (the hands are the classic example, though that's getting harder to rely on than it used to be).

Combine all of that with our knowledge of Second Life™, how it looks and what's actually achievable in-world, and we catch the large majority of content that's been generated or altered by AI. But we won't pretend we catch everything. By definition, the cases we miss are the ones we didn't spot. Some sophisticated edits will slip through, and the report system helps us give attention to those posts. But it may mean that it’s close enough to be a Second Life™ picture, and we are ok with accepting this because we cannot confirm it’s generative AI.

The hardest edge case is products that use AI-enhanced or AI-generated textures, hair especially. There, our job is to figure out whether the altered texture is part of the product or something the user added themselves. To do that, we may look for the original product or go in-world to check.

Rising AI tools

Primfeed's AI rules are about the end result and the overall look of the picture. Using AI tools to erase a background or change the color of your shirt is fine. What we don't want to see is altered or generated textures, hair, clothing, and so on.

Dealing with the human

This part is tricky, because there are several sides to it. Some users don't understand what counts as AI-generated content, some don't realize the tool they used relies on generative AI, and occasionally someone insists their picture is untouched when it isn't.

To be as accurate as possible, when multiple moderators have a doubt about whether someone used generative AI on their pictures, we may ask for a screenshot of your the external editing software and layers or the original raw picture to compare against the end result.

While we would love to trust everyone, sometimes the layers tell a different story than what we were told. Here's an example: the picture below is a stack of Photoshop layers. Layer 3 is the original raw picture from Second Life™, and layers 1 and 2 are AI-enhanced parts of layer 3, applied with a mask. So the image is based on the raw Second Life™ picture, but small AI-enhanced bits have been blended in, something that isn't visible from the final result alone.

This is exactly why we ask for the layers or the raw picture. It lets us see whether something was painted, drawn, or AI-enhanced, and it shows us what really happened. This may sound intrusive but this is the only way we have at this time to confirm or rule out our doubts and residents trust us to do a proper moderation job.

Balancing different opinions, tastes, and sensibilities

Second Life™ is an international social platform, which means different cultures and different sensibilities about content, and about whether one post is adult and another is general.

We try to keep Primfeed’s Terms of Service as simple as possible and close to Second Life™ when it comes to Maturity Rating, with a help page to better guide you: Maturity Rating.

To make moderation easier, we've decided to draw strict lines around certain types of content. We agree that, depending on context and on how much a picture actually "shows", some pictures could arguably have a lower rating, but allowing for that introduces a lot of complexity.

Our team members come from Canada, the US, the UK, and France, so we have a wide range of opinions, which helps us find the middle ground on the rules and the actions we take around ratings.

What are the risks if I get moderated?

It depends entirely on what you did "wrong" and how often. For example, if you post a thin bikini that shows pubic hair under the General rating once, your post will simply be corrected to Moderate by the moderators and you don't risk anything.

However, if you keep making the same mistakes, we may temporarily restrict your account. The usual duration is 1 day, but it can be 3, 7, 14 days, and so on, depending on how many restrictions you've already had in the past. Note that if you reach out and understand why you were restricted in the first place, we're happy to lift the restriction before it expires.

If you post an Adult+ picture under the General rating, that's serious, and you may be restricted for one or more days right away.

We don't think permanently suspending an account is a good solution. But if moderation notifications are repeatedly ignored, we may end up doing it, to protect the other users and be fair to them.

What if you think we got it wrong?

The moderation team is human, sometimes we'll misjudge a post. If you think we made the wrong call on your content, tell us. The fastest way to clear things up is to send us the raw picture or the layers so we can compare for ourselves. If we got it wrong, we'll fix it. You can reach out to PrimfeedSupport Resident in-world.

Why don’t you reach out before taking any actions?

Every day, there are 5,000 media posted on Primfeed and we take around 200 actions per day. While we agree that it’s the best solution to reach out before, sending ~100-200 messages per day, and waiting for an answer isn’t the best moderation reaction possible.

However after feedback from the community, the appeal process will be shown clearly on the app, so it’s fast to contact us when needed.

Rules choices

Allowing generative AI under conditions for stores

The current rules allow commercial content to use generative AI, with one specific condition: If the product has been generated or altered by generative AI, the label AI must be added to the post.

The thinking behind allowing stores to use AI is that, at the time of launch, Primfeed needed everyone on board, and in the months and years before launch, stores had been using generative AI or morph techniques to promote and sell their products in Second Life™.

So as a middle ground, we chose to allow it under the AI label, so that (1) stores can still be part of Primfeed, and (2) customers are aware that what they're buying may look different in-world. Discussions are ongoing to see if eventually, we will disallow completely generative AI for everyone.

DJ events are not allowed in the Public Gallery

Primfeed doesn't have any algorithm yet, so every feed is ordered newest-first. This is useful and fair to everyone, because everyone has an equal chance to have their work noticed, but it also means that the more content there is, the further down the list your content goes quickly.

DJ events represent a lot of content, so we've decided not to allow them in the Public Gallery, to prevent the daily DJ events from spamming it. They can still be created under the dedicated category: Events.

Why the Public Gallery has more rules?

The Public Gallery is Primfeed’s home page, and it’s the best way to promote your work on Primfeed. Since there is no algorithm yet, we wanted to have the cleanest feed possible, hence the extra rules.

Beyond that, it allows everyone to have a small bit of non Second Life™ content on their profile, around 10 - 20%.

One final word

Primfeed is a small team trying to build something different in today's internet. We try as hard as we can to make something that feels authentic and refreshing, built around something we all love: Second Life™.

When Primfeed launched, I, Luke, wanted to propose something unique: limiting the use of generative AI on the platform. Most online platforms allow any kind of content, and that's their choice. But I wanted to try a different path, one that puts Second Life™ and the craft behind every picture first.

Generative AI doesn't represent what Second Life™ is about: creativity, humans, and social bonds.

Thank you for being so supportive over the years, and please always keep sending feedback in the portal for us to know what to improve: https://feedback.primfeed.com/