Lightnews — Scholar-powered news

Aaron Rodericks @aaron.bsky.team · 3h

I'll be trying out leaflet.pub moving forward to try to write longer form blogs on Trust and Safety so that I can explain things in a bit more depth.
leaflet.pub/25030fdf-2a8...

Context Collapse & T&S

One of the hardest things about writing on a microblogging service is context collapse, especially when it comes to complex issues — you reply to one person, others take it out of context, and suddenl...

leaflet.pub

10 8 53

Aaron Rodericks @aaron.bsky.team · 5h

Tbh I believe most of your concerns are with content moderation as a system. Some of this can be improved as we mature as a tiny startup, but fundamentally it would take a longer explanation than I can cover in posts to cover why it’s so difficult to make decisions at scale. I’ll send a ping.

2 2

Reposted by Aaron Rodericks

Bluesky @bsky.app · Nov 15

Bluesky uses AI internally to assist in content moderation, which helps us triage posts and shield human moderators from harmful content. We also use AI in the Discover algorithmic feed to serve you posts that we think you’d like.

None of these are Gen AI systems trained on user content.

350 3.2K 35K

Aaron Rodericks @aaron.bsky.team · 16h

I wish you good luck!

2 14

Aaron Rodericks @aaron.bsky.team · 18h

Every appeal gets reviewed by moderators, but literally no social media site at scale can scan user content in a timely fashion for violations using only humans. They do not retain or train on any Bsky user data or images.

13 15 52

Aaron Rodericks @aaron.bsky.team · 21h

thehive.ai evaluates all images posted to the site. We don't actually provide any input.

AI to Understand, Search, and Generate Content | Hive

Hive's APIs enable developers to integrate pre-trained AI models that address technically challenging content understanding needs into their applications.

thehive.ai

190 34 45

Aaron Rodericks @aaron.bsky.team · 1d

It's an automated system, so unfortunately this is an error that gets made surprisingly often with prosthetics.

79 6 28

Aaron Rodericks @aaron.bsky.team · 1d

the most recent one was a temp error on asukafield where a mod hid the list and it was rapidly reversed on invesigation. Is that the one you are referencing, or just lists in general?

4 5

Aaron Rodericks @aaron.bsky.team · 1d

It was a brand new bot that we had just activated to deal with the massive backlog in reports after the US election.

We currently have a much smaller backlog and mods are behind on handling reports, since we had a significant surge in reports that we are still catching up on.

1 6

Aaron Rodericks @aaron.bsky.team · 1d

TBH that was an error on our end that I wasn't aware of until extremely recently pfrazee.leaflet.pub/3lz4sgu7iec2k. It's being addressed architecturally. As far as I'm aware we only ban CSAM at a relay level, but I'm not particularly technical.

Update on Protocol Moderation - Paul's Leaflets

Where account takedowns happen is important

pfrazee.leaflet.pub

1 5

Aaron Rodericks @aaron.bsky.team · 1d

yes we had an impersonation bot that assumed the account was fake upon signup (because people had created similar fake account prior so it focused on that pattern). That was totally unrelated to any assessment under the community guidelines.

1 7

Aaron Rodericks @aaron.bsky.team · 1d

That wording is in our community guidelines. bsky.social/about/suppor... It's still on our site and they expire on Oct 15 2025. So seeing as the wording is still live and on the site, how were they changed last December?

Community Guidelines (Deprecated) - Bluesky

bsky.social

2 4

Aaron Rodericks @aaron.bsky.team · 1d

cause people raise it in my mentions daily and literally won't stop, so I might as well share my side.

26 16

Aaron Rodericks @aaron.bsky.team · 1d

We know it’s frustrating to see bad behaviour elsewhere. But trying to police the entire internet isn’t possible for any team — and making exceptions about when to do so only adds confusion and inconsistency. 9/9

87 2 92

Aaron Rodericks @aaron.bsky.team · 1d

This approach keeps our system scalable and fair as we grow. Every decision must be explainable, repeatable, and based on evidence our tools can reliably evaluate. We’re also working to make our decisions clearer and more transparent. 8/9

8 1 69

Aaron Rodericks @aaron.bsky.team · 1d

At Bluesky, moderation focuses on behaviour we can observe here: harassment, threats, hate, spam, impersonation, and other violations of our Community Guidelines. 7/9

21 6 72

Aaron Rodericks @aaron.bsky.team · 1d

When moderation expands into other apps or private servers, we risk huge workloads, privacy issues, and inconsistent enforcement. That’s why nearly every platform limits enforcement to what happens on their own service. 6/9

3 4 80

Aaron Rodericks @aaron.bsky.team · 1d

Screenshots can be edited, context can be missing, and identities are hard to confirm across platforms. Moderation depends on repeatable, auditable processes — not assumptions based on unverifiable off-platform material. 5/9

8 4 80

Aaron Rodericks @aaron.bsky.team · 1d

Let’s say someone sends us screenshots or Discord logs — 15,000 lines of conversation — claiming they prove coordinated harassment (this happened). That’s hours or days of work to verify and still might not show any clear link to behaviour on Bluesky. 4/9

9 5 86

Aaron Rodericks @aaron.bsky.team · 1d

Right now, we’re a small team with a growing backlog. The slower we get, the more people wonder why decisions aren’t being made in real time — but with human review and millions of posts, that’s simply not possible. 3/9

18 1 72

Aaron Rodericks @aaron.bsky.team · 1d

A single moderator reviews thousands of reports per year — often making a decision in under 30 seconds. Expanding the scope of what counts as “evidence” makes it much harder to act fairly and consistently. 2/9

13 79

Aaron Rodericks @aaron.bsky.team · 1d

People often ask why we don’t act on off-platform evidence when investigating reports. Here’s an example of how complex that can get. 1/9

100 23 150

Aaron Rodericks @aaron.bsky.team · 1d

That was always the case? I'll post a longer thread explaining why.

9 3

Aaron Rodericks @aaron.bsky.team · 1d

that second was a mod error and she has been released with an apology already. Mods make errors continually with hundred mods reviewing tickets in 30 seconds or less.

9 6

Aaron Rodericks @aaron.bsky.team · 1d

If you want to have a conversation about it, or other mod decisions that you disagree with, sure - I can understand that, but endlessly coming up with a screenshot as evidence of...? isn't particularly constructive.

31 24