Aaron Rodericks

Ari Cohn @aricohn.com [ View ] Jul 01, 2024 at 14:41 UTC

This is it. This is the point.

1 replies 6 reposts 30 likes

Aaron Rodericks @aaron.bsky.team [ View ] Jun 29, 2024 at 17:42 UTC

Oh I know nothing technical.

0 replies 0 reposts 2 likes

Aaron Rodericks @aaron.bsky.team [ View ] Jun 29, 2024 at 13:28 UTC

Believe I wrote one up when I joined, but I haven't done any public documentation. Will discuss internally. This for development of a client app?

2 replies 0 reposts 1 likes

Aaron Rodericks @aaron.bsky.team [ View ] Jun 29, 2024 at 13:01 UTC

When accounts make a number of similar infractions they get an account-level label that shows up on every post. That account has one on every post. In the future we hope to clarify those labels to avoid confusion.

1 replies 0 reposts 3 likes

Aaron Rodericks @aaron.bsky.team [ View ] Jun 29, 2024 at 11:48 UTC

It was designed so that you don't lose existing convos should you change your settings. I believe if you leave and/or delete the convo then the initiation logic kicks back in, but I haven't had coffee yet.

1 replies 0 reposts 7 likes

Aaron Rodericks @aaron.bsky.team [ View ] Jun 28, 2024 at 18:13 UTC

Best way to address is for them to check the emails and to respond to moderation@

0 replies 0 reposts 2 likes

Aaron Rodericks @aaron.bsky.team [ View ] Jun 28, 2024 at 17:20 UTC

They should have received an email informing them of the rationale for the label. I checked and two warnings were issued.

2 replies 0 reposts 2 likes

Aaron Rodericks @aaron.bsky.team [ View ] Jun 28, 2024 at 15:34 UTC

Can certainly look into it.

1 replies 0 reposts 4 likes

Aaron Rodericks @aaron.bsky.team [ View ] Jun 28, 2024 at 14:56 UTC

I have a rough counter in my job where I see how many days I can go without someone mentioning a bloom filter, and was just on my longest streak to date. Been thinking of recently adding roaring bitmaps to that terminology list.

1 replies 0 reposts 9 likes

Aaron Rodericks @aaron.bsky.team [ View ] Jun 28, 2024 at 14:44 UTC

Understood - there are limitations around lists, so don't believe we'd want to drive more adoption of that feature. ie once you go above 5k accounts listed, it breaks things.

3 replies 0 reposts 0 likes

Aaron Rodericks @aaron.bsky.team [ View ] Jun 28, 2024 at 14:30 UTC

not entirely sure I understand, but making some changes to self-labels that may assist, and we can make that more flexible in time

1 replies 0 reposts 2 likes

Aaron Rodericks @aaron.bsky.team [ View ] Jun 28, 2024 at 01:37 UTC

Np

1 replies 0 reposts 2 likes

Aaron Rodericks @aaron.bsky.team [ View ] Jun 28, 2024 at 00:54 UTC

will add you to enhanced defences if you aren't already.

1 replies 0 reposts 8 likes

Aaron Rodericks @aaron.bsky.team [ View ] Jun 28, 2024 at 00:35 UTC

To my limited understanding it's not fully implemented.

0 replies 0 reposts 1 likes

Aaron Rodericks @aaron.bsky.team [ View ] Jun 27, 2024 at 23:36 UTC

There are changes to self labels on the roadmap. Hopefully soon.

0 replies 0 reposts 8 likes

Aaron Rodericks @aaron.bsky.team [ View ] Jun 27, 2024 at 18:05 UTC

Depends on what you mean. Ie all surfaces are rate limited, but then the policy threshold is usually lower in terms of when we issue warnings / take action / etc...

1 replies 0 reposts 3 likes

Aaron Rodericks @aaron.bsky.team [ View ] Jun 27, 2024 at 18:00 UTC

it's more of 'huh, that was unexpected'. Conversations around product launches, such as your concerns are also valuable and feed into the signal - since it's not just T&S saying these things. If anything, thanks for raising your concerns. Those + abuse signals help us design a better Bluesky.

1 replies 0 reposts 17 likes

Aaron Rodericks @aaron.bsky.team [ View ] Jun 27, 2024 at 17:58 UTC

Don't let me convince you that I'm an all-seeing, all-knowing oracle. I am not, despite having done this for a while. The design process always involves tensions and trade-offs. In some cases it's a 'I told you so'... okay, frequently it's a 'I told you so', but every now and again..

1 replies 0 reposts 12 likes

Aaron Rodericks @aaron.bsky.team [ View ] Jun 27, 2024 at 17:49 UTC

There are a certain amount of mitigations baked in at launch, and if we get signal that an abuse vector is higher than we expected, or that we didn't predict abuse vectors accurately, then I'll have enough evidence that we need more resources for product mitigations sooner rather than later.

2 replies 0 reposts 19 likes

Aaron Rodericks @aaron.bsky.team [ View ] Jun 27, 2024 at 17:46 UTC

Part of the issue is the way the underlying protocol is designed makes certain things like an opt out function a heavy lift. I hear you, and that's why for all new product launches has pre-launch, post-launch, and downstream mitigations.

1 replies 1 reposts 12 likes

Aaron Rodericks @aaron.bsky.team [ View ] Jun 27, 2024 at 17:41 UTC

Plenty in the design phase, but a number of my concerns have been mitigated pre-launch via guardrails in the product. Ie max amount of users on list, who can use them, who can create them, how they are flagged to mods, how easy to action etc...

1 replies 0 reposts 5 likes

Aaron Rodericks @aaron.bsky.team [ View ] Jun 27, 2024 at 17:11 UTC

That would be me.

3 replies 1 reposts 23 likes

Aaron Rodericks @aaron.bsky.team [ View ] Jun 27, 2024 at 17:06 UTC

People really tend to over index on low frequency and high severity. Just due to how we process danger. Ie terrorism. vs the reality of how many kids die annually from hot dogs or marshmallows.

1 replies 0 reposts 11 likes

Aaron Rodericks @aaron.bsky.team [ View ] Jun 27, 2024 at 17:01 UTC

The longer version of which we refer to in trust and safety as a harms analysis. Usually measured on prevalence vs severity, but you can factor in other risks. Comes down to: How frequently is X bad thing going to happen? VS How much harm occurs?

1 replies 1 reposts 7 likes

Aaron Rodericks @aaron.bsky.team [ View ] Jun 27, 2024 at 16:57 UTC

These would fall under thought exercises which is the first step of ‘these are all the things that could go wrong’. That’s where most people are in the responses. What’s missing is, how likely is this to occur? Followed by - how much evidence is there that this is happening?

1 replies 1 reposts 9 likes

Aaron Rodericks @aaron.bsky.team [ View ] Jun 27, 2024 at 15:37 UTC

As a small startup, if we get signal, we can pivot rapidly to adjust. It's preferable to over-engineering product for every theoretical abuse vector, and have no one use it (that's how things are built in govt). I'll review all reports for starter packs T+7/T+14/T+30 and see what the data shows.

10 replies 1 reposts 7 likes

Aaron Rodericks @aaron.bsky.team [ View ] Jun 27, 2024 at 15:14 UTC

That's not to say that we've mitigated every risk, and human ingenuity never fails to amaze. If we see these becoming a vector for abuse, we'll prioritize changes to the product to add elements such as more user controls.

4 replies 1 reposts 11 likes

Aaron Rodericks @aaron.bsky.team [ View ] Jun 27, 2024 at 14:47 UTC

We do red team out new product features/surfaces, and I believe that some of these risks are mitigated, ie these are only for new users to join and follow these specific users. They are also fully integrated into reporting/mod review for cases of potential harassment.

2 replies 1 reposts 11 likes

Aaron Rodericks @aaron.bsky.team [ View ] Jun 27, 2024 at 13:01 UTC

I use @mackuba.eu's list - blue.mackuba.eu/labellers/ but you can also use labeller feeds/lists - bsky.app/profile/psin... or alternatively searching for the term labeler will bring up a number.

1 replies 0 reposts 5 likes

Aaron Rodericks @aaron.bsky.team [ View ] Jun 27, 2024 at 12:41 UTC

Oh yes, that was a good reminder for me to subscribe to the IFTAS labeller.

1 replies 0 reposts 2 likes

Aaron Rodericks @aaron.bsky.team [ View ] Jun 27, 2024 at 12:26 UTC

Yup, and I need to do a lot better at communicating what those lines are, but we need better back end architecture, messaging, user comms, etc... so that it's clear on getting a warning / label / takedown etc what exactly the issue was, but that takes a lot of structural back-end work.

1 replies 3 reposts 45 likes

Aaron Rodericks @aaron.bsky.team [ View ] Jun 27, 2024 at 12:24 UTC

TBH, I shifted things much more towards takedowns perspective on things like clear spam and scams. That meant evolutions in our label usage. For example - our spam label is used now for when people are potentially well-intentioned, and their behaviour can be corrected.

1 replies 1 reposts 23 likes

Aaron Rodericks @aaron.bsky.team [ View ] Jun 27, 2024 at 12:21 UTC

I assure you we don't even prioritize reports from you! (though report schema prioritization is on my to do list - ie bad things get surfaced faster). Need to pull better stats, but nearly all reports get reviewed and actioned in under 24 hours.

2 replies 0 reposts 22 likes

Aaron Rodericks @aaron.bsky.team [ View ] Jun 27, 2024 at 12:18 UTC

awesome, thanks! My long reads here are all academic papers so this will be comparatively a lot of fun.

0 replies 0 reposts 4 likes

Aaron Rodericks @aaron.bsky.team [ View ] Jun 27, 2024 at 12:13 UTC

If Bluesky/atproto continues to grow (and that's indeed why I'm here!) part of the promise is to have account portability that works across apps on the ecosystem, so even for our own labeller, we'd want people to have their labels/bans roll off depending on the policy violation.

1 replies 3 reposts 21 likes

Aaron Rodericks @aaron.bsky.team [ View ] Jun 27, 2024 at 12:09 UTC

I'm coming more from the perspective of restorative justice, in that people spend their digital lives online, and sometimes things get heated, have episodes, or do things as teens. Should bans or labels follow then for years or decades? Twitter is getting almost to 20 years old!

1 replies 2 reposts 18 likes

Aaron Rodericks @aaron.bsky.team [ View ] Jun 27, 2024 at 11:33 UTC

This is an evolutionary experiment. The first stage already showed us a range of great possibilities that I certainly didn’t have in mind as use cases.

0 replies 1 reposts 7 likes

Aaron Rodericks @aaron.bsky.team [ View ] Jun 27, 2024 at 11:26 UTC

This book www.goodreads.com/book/show/36... or this one? www.goodreads.com/book/show/60...

1 replies 0 reposts 1 likes