Ryan J. Gallagher
@ryanjgallag.com
Applied scientist trying to make the internet a little better. PhD. Trust & safety, platform manipulation, networks, fingerstyle guitar. I use my hair to express myself. He/they
Pinned
Ryan J. Gallagher
@ryanjgallag.com
· Nov 12
With more people migrating to Bluesky (and my general quietness since Twitter fell apart), I want to reintroduce myself!
I'm an applied scientist in online trust & safety. I specialize in detecting coordinated/harmful/deceptive networks, and bridging domain expertise with machine learning systems
I'm an applied scientist in online trust & safety. I specialize in detecting coordinated/harmful/deceptive networks, and bridging domain expertise with machine learning systems
Reposted by Ryan J. Gallagher
I need everybody to read this and understand what’s going on. These are spouses of US citizens in the country legally. They have entry clearance and green cards, they’ve passed background checks and have no criminal records. ICE is taking them anyway.
Gift link:
www.nytimes.com/2025/11/26/u...
Gift link:
www.nytimes.com/2025/11/26/u...
Green Card Interviews End in Handcuffs for Spouses of U.S. Citizens
www.nytimes.com
November 26, 2025 at 8:40 PM
I need everybody to read this and understand what’s going on. These are spouses of US citizens in the country legally. They have entry clearance and green cards, they’ve passed background checks and have no criminal records. ICE is taking them anyway.
Gift link:
www.nytimes.com/2025/11/26/u...
Gift link:
www.nytimes.com/2025/11/26/u...
Reposted by Ryan J. Gallagher
some might know how @roost.tools is approaching open safety tool development, but many more may not! @annebdh.bsky.social lays it all out here, including our governance structure, how people can get involved, and how working in the open doesn't mean more abuse!
roost.tools/blog/open-by...
roost.tools/blog/open-by...
Open by Design: ROOST's Approach to Safety Tool Development
Robust Open Online Safety Tools or ROOST is a new non-profit entity designed to address the urgent need for accessible, high-quality safety tools in the rapidly evolving digital landscape.
roost.tools
November 26, 2025 at 5:23 PM
some might know how @roost.tools is approaching open safety tool development, but many more may not! @annebdh.bsky.social lays it all out here, including our governance structure, how people can get involved, and how working in the open doesn't mean more abuse!
roost.tools/blog/open-by...
roost.tools/blog/open-by...
Reposted by Ryan J. Gallagher
a light thread on what we mean by "user engagement". 🧵 1/26
November 25, 2025 at 4:07 PM
a light thread on what we mean by "user engagement". 🧵 1/26
Reposted by Ryan J. Gallagher
“…a major advance that we’ve been trying to achieve for a very long time, and a great aid to scholarship. It allows human beings to focus their time on the important, profound work of understanding another human being, rather than staring at a curlicue to grasp if it’s an L or an I.”
New issue of my newsletter: "The Writing Is on the Wall for Handwriting Recognition" — One of the hardest problems in digital humanities has finally been solved, and it's a good use of AI newsletter.dancohen.org/archive/the-...
The Writing Is on the Wall for Handwriting Recognition
One of the hardest problems in digital humanities has finally been solved
newsletter.dancohen.org
November 26, 2025 at 1:02 AM
“…a major advance that we’ve been trying to achieve for a very long time, and a great aid to scholarship. It allows human beings to focus their time on the important, profound work of understanding another human being, rather than staring at a curlicue to grasp if it’s an L or an I.”
Reposted by Ryan J. Gallagher
It’s blowing my mind that schools, universities, public services would run headlong into this. We spent 15 years documenting black boxes. This is a black box in a black hole!
I don't understand how anyone can watch how blatantly Grok is manipulated to answer the way ownership desires it to and then act like the other LLM chatbots couldn't possibly be similarly but less obviously compromised to produce responses in whatever way corporate interests and priorities dictate.
November 23, 2025 at 11:16 PM
It’s blowing my mind that schools, universities, public services would run headlong into this. We spent 15 years documenting black boxes. This is a black box in a black hole!
Reposted by Ryan J. Gallagher
We're announcing a new partnership with StopNCII.org to prevent non-consensual intimate imagery (NCII) from spreading on Bluesky. While NCII hasn't been frequent on our platform, these incidents are devastating for victims and preventing them is a top safety priority. 1/4
Stop Non-Consensual Intimate Image Abuse | StopNCII.org
StopNCII.org is operated by the Revenge Porn Helpline which is part of SWGfL, a charity that believes that all should benefit from technology, free from harm.
StopNCII.org
November 25, 2025 at 5:29 PM
We're announcing a new partnership with StopNCII.org to prevent non-consensual intimate imagery (NCII) from spreading on Bluesky. While NCII hasn't been frequent on our platform, these incidents are devastating for victims and preventing them is a top safety priority. 1/4
Reposted by Ryan J. Gallagher
If everything is a top priority, then everything gets done quickly. That's basic management
November 25, 2025 at 8:15 PM
If everything is a top priority, then everything gets done quickly. That's basic management
Reposted by Ryan J. Gallagher
Social media researcher @taliastroud.bsky.social (New_ Public and @engagingnews.bsky.social) presents our new nationwide poll results on how Americans really connect with their neighbors online — and the surprising gaps that deserve more attention.
Watch here 📹
Watch here 📹
The Local Connection Crisis: New Data on What Communities Need
YouTube video by New_ Public
youtu.be
November 25, 2025 at 6:42 PM
Social media researcher @taliastroud.bsky.social (New_ Public and @engagingnews.bsky.social) presents our new nationwide poll results on how Americans really connect with their neighbors online — and the surprising gaps that deserve more attention.
Watch here 📹
Watch here 📹
Reposted by Ryan J. Gallagher
Here it is, the most damning academic figure in the history of all figures: Russian IRA accounts deeply embedded in both sides of the 2016 Twitter BLM "discussion" (by @katestarbird.bsky.social & team, 2018).
THIS is what our information ecosystem supports. Twitter just made it 1% more visible.
THIS is what our information ecosystem supports. Twitter just made it 1% more visible.
November 25, 2025 at 1:30 PM
Here it is, the most damning academic figure in the history of all figures: Russian IRA accounts deeply embedded in both sides of the 2016 Twitter BLM "discussion" (by @katestarbird.bsky.social & team, 2018).
THIS is what our information ecosystem supports. Twitter just made it 1% more visible.
THIS is what our information ecosystem supports. Twitter just made it 1% more visible.
Why do most AI (ie data science and machine learning) roadmaps fail?
Because leadership wants the magic payoff without being told there are steps outside just training a model
Because leadership wants the magic payoff without being told there are steps outside just training a model
Why do most AI roadmaps fail? 🫠
And no, it's not "messy data"
theimpostor.substack.com
November 25, 2025 at 11:30 AM
Why do most AI (ie data science and machine learning) roadmaps fail?
Because leadership wants the magic payoff without being told there are steps outside just training a model
Because leadership wants the magic payoff without being told there are steps outside just training a model
Reposted by Ryan J. Gallagher
X’s New Feature Reveals Why Trust & Safety Work Was Never About The ‘Censorship Industrial Complex’
For the last few years, Matt Taibbi, Michael Shellenberger, and their allies have spent considerable energy attacking both academic researchers studying disinformation and the trust & safety teams…
For the last few years, Matt Taibbi, Michael Shellenberger, and their allies have spent considerable energy attacking both academic researchers studying disinformation and the trust & safety teams…
X’s New Feature Reveals Why Trust & Safety Work Was Never About The ‘Censorship Industrial Complex’
For the last few years, Matt Taibbi, Michael Shellenberger, and their allies have spent considerable energy attacking both academic researchers studying disinformation and the trust & safety teams at social media platforms working to identify and remove coordinated inauthentic behavior—particularly foreign influence operations. They've insisted that any attempt to study and limit such operations is actually just "censorship" with various forms of cover, whether academic or operational.
www.techdirt.com
November 24, 2025 at 5:25 PM
X’s New Feature Reveals Why Trust & Safety Work Was Never About The ‘Censorship Industrial Complex’
For the last few years, Matt Taibbi, Michael Shellenberger, and their allies have spent considerable energy attacking both academic researchers studying disinformation and the trust & safety teams…
For the last few years, Matt Taibbi, Michael Shellenberger, and their allies have spent considerable energy attacking both academic researchers studying disinformation and the trust & safety teams…
Reposted by Ryan J. Gallagher
Many people don't know this but arXiv is maintained and operated by @cornelltech.bsky.social (that does NOT mean you can send me complaints or feature requests!)
Everyone in the community should be excited that arXiv got $7M to upgrade from Schmidt Sciences, NASA
news.cornell.edu/stories/2025...
Everyone in the community should be excited that arXiv got $7M to upgrade from Schmidt Sciences, NASA
news.cornell.edu/stories/2025...
$7M grant from NASA, Schmidt Sciences to upgrade arXiv | Cornell Chronicle
The funding will help arXiv – which is maintained and operated by Cornell Tech – finish migrating to cloud infrastructure and modernizing its code.
news.cornell.edu
November 24, 2025 at 9:32 PM
Many people don't know this but arXiv is maintained and operated by @cornelltech.bsky.social (that does NOT mean you can send me complaints or feature requests!)
Everyone in the community should be excited that arXiv got $7M to upgrade from Schmidt Sciences, NASA
news.cornell.edu/stories/2025...
Everyone in the community should be excited that arXiv got $7M to upgrade from Schmidt Sciences, NASA
news.cornell.edu/stories/2025...
Reposted by Ryan J. Gallagher
Lots of people are calling out the fact that X is filled with fake MAGA influencers from far away countries, but I wanted to point out that this, not "ideological censorship," is why companies have trust & safety teams that work on stopping inauthentic behavior.
X’s New Feature Reveals Why Trust & Safety Work Was Never About The ‘Censorship Industrial Complex’
For the last few years, Matt Taibbi, Michael Shellenberger, and their allies have spent considerable energy attacking both academic researchers studying disinformation and the trust & safety teams…
For the last few years, Matt Taibbi, Michael Shellenberger, and their allies have spent considerable energy attacking both academic researchers studying disinformation and the trust & safety teams…
X’s New Feature Reveals Why Trust & Safety Work Was Never About The ‘Censorship Industrial Complex’
For the last few years, Matt Taibbi, Michael Shellenberger, and their allies have spent considerable energy attacking both academic researchers studying disinformation and the trust & safety teams at social media platforms working to identify and remove coordinated inauthentic behavior—particularly foreign influence operations. They've insisted that any attempt to study and limit such operations is actually just "censorship" with various forms of cover, whether academic or operational.
www.techdirt.com
November 24, 2025 at 5:57 PM
Lots of people are calling out the fact that X is filled with fake MAGA influencers from far away countries, but I wanted to point out that this, not "ideological censorship," is why companies have trust & safety teams that work on stopping inauthentic behavior.
Reposted by Ryan J. Gallagher
The UW Center for an Informed Public is looking for postdocs (for 2026-2028) from across diverse disciplines whose research sheds light on the challenges of our modern information environment, promotes civic health, and/or helps people/communities navigate online spaces: apply.interfolio.com/177901
Apply - Interfolio
{{$ctrl.$state.data.pageTitle}} - Apply - Interfolio
apply.interfolio.com
November 24, 2025 at 6:26 PM
The UW Center for an Informed Public is looking for postdocs (for 2026-2028) from across diverse disciplines whose research sheds light on the challenges of our modern information environment, promotes civic health, and/or helps people/communities navigate online spaces: apply.interfolio.com/177901
Reposted by Ryan J. Gallagher
Our postdoc search is open! socialmediacollective.org/2025/11/20/s...
Apply for our sociotechnical systems postdoc position, for July 2026
Microsoft Research New England is looking for a postdoctoral researcher interested in bringing sociotechnical perspectives to analyze critical issues of our time. They will join a team of social sc…
socialmediacollective.org
November 20, 2025 at 7:35 PM
Our postdoc search is open! socialmediacollective.org/2025/11/20/s...
Reposted by Ryan J. Gallagher
Latest out in Plos Comp Bio: "Epydemix: An open-source Python package for epidemic modeling with integrated approximate Bayesian calibration". A new package in Python for epidemic modeling. Paper: journals.plos.org/ploscompbiol.... Website with lots of tutorials: www.epydemix.org
November 20, 2025 at 7:59 AM
Latest out in Plos Comp Bio: "Epydemix: An open-source Python package for epidemic modeling with integrated approximate Bayesian calibration". A new package in Python for epidemic modeling. Paper: journals.plos.org/ploscompbiol.... Website with lots of tutorials: www.epydemix.org
Reposted by Ryan J. Gallagher
We also provided the first-ever full data archive of @bsky.app complete with hydrated objects at the point of ingest alongside a dozen ML classifications, for the cheapest possible rate, to other vetted builders and researchers in the ecosystem
Announcing the Graze Archives
A brief tour of the S3 requestor-pays archives of the Graze turbostream and freshly-announced megastream
www.graze.social
November 22, 2025 at 7:59 PM
We also provided the first-ever full data archive of @bsky.app complete with hydrated objects at the point of ingest alongside a dozen ML classifications, for the cheapest possible rate, to other vetted builders and researchers in the ecosystem
Reposted by Ryan J. Gallagher
Milestone alert: We're celebrating our first anniversary today! In just one year, we went from an idea on a blog post to sending 8.6 *million* users 23.5 *billion* posts across thousands of feeds created by thousands of brilliant, creative people who are working together to create a better internet.
November 22, 2025 at 7:59 PM
Milestone alert: We're celebrating our first anniversary today! In just one year, we went from an idea on a blog post to sending 8.6 *million* users 23.5 *billion* posts across thousands of feeds created by thousands of brilliant, creative people who are working together to create a better internet.
Reposted by Ryan J. Gallagher
We've got our follow-up on our deep dive into Rumble, this time looking at the coordinated networks of scammers targeting fans of the platform's top creators with impersonation scams. The scale of these efforts is really insane to wrap your head around.
openmeasures.io/dan-bongino-...
openmeasures.io/dan-bongino-...
Impersonation Scams Target Audiences of Rumble Creators - Open Measures
Open Measures builds accessible and transparent products to contextualize the spread of harmful online information that impacts public discourse and global events.
openmeasures.io
November 21, 2025 at 11:24 PM
We've got our follow-up on our deep dive into Rumble, this time looking at the coordinated networks of scammers targeting fans of the platform's top creators with impersonation scams. The scale of these efforts is really insane to wrap your head around.
openmeasures.io/dan-bongino-...
openmeasures.io/dan-bongino-...
Reposted by Ryan J. Gallagher
SVMs are the only moral type of machine learning, which is why they’re still taught in ML classes. Unlike next token prediction, they are inherently good
November 21, 2025 at 9:02 PM
SVMs are the only moral type of machine learning, which is why they’re still taught in ML classes. Unlike next token prediction, they are inherently good
Reposted by Ryan J. Gallagher
Today on @indicator.media's (free) Briefing: A twinkle of optimism on the AI slop front as TikTok and Google roll out tools to filter and detect AI content.
I remain bearish about generative AI and info quality. But I’ll celebrate developments that increase user agency and platform transparency.
I remain bearish about generative AI and info quality. But I’ll celebrate developments that increase user agency and platform transparency.
Briefing: Good news on the AI slop front. No, really!
Plus: detecting AI-generated images with SynthID, and advanced archival OSINT
indicator.media
November 21, 2025 at 12:54 PM
Today on @indicator.media's (free) Briefing: A twinkle of optimism on the AI slop front as TikTok and Google roll out tools to filter and detect AI content.
I remain bearish about generative AI and info quality. But I’ll celebrate developments that increase user agency and platform transparency.
I remain bearish about generative AI and info quality. But I’ll celebrate developments that increase user agency and platform transparency.
Reposted by Ryan J. Gallagher
The underlying study: "Experimental evidence of the effects of large language models versus web search on depth of learning" academic.oup.com/pnasnexus/ar...
"participants reported developing shallower knowledge from LLM summaries even when the results were augmented by...web links" vs. search
"participants reported developing shallower knowledge from LLM summaries even when the results were augmented by...web links" vs. search
November 21, 2025 at 1:31 PM
The underlying study: "Experimental evidence of the effects of large language models versus web search on depth of learning" academic.oup.com/pnasnexus/ar...
"participants reported developing shallower knowledge from LLM summaries even when the results were augmented by...web links" vs. search
"participants reported developing shallower knowledge from LLM summaries even when the results were augmented by...web links" vs. search
Reposted by Ryan J. Gallagher
Once again astonished that Elon is building a propaganda machine in public, saying "hello, here is my propaganda machine", and the tech community continues to use it
November 21, 2025 at 12:06 AM
Once again astonished that Elon is building a propaganda machine in public, saying "hello, here is my propaganda machine", and the tech community continues to use it
Reposted by Ryan J. Gallagher
If you're trying to convince your friends to spend time on Bluesky rather than X, maybe show them this. X is not "the public town square" or about "free speech." It's "just what Elon Musk wants you to see." If you see stuff he doesn't like he'll step in to make sure you don't see it for long.
If you scroll on X, you are only being shown what Elon wants you to see. If you create content on X, no one will see it unless Elon wants them to.
November 21, 2025 at 1:31 AM
If you're trying to convince your friends to spend time on Bluesky rather than X, maybe show them this. X is not "the public town square" or about "free speech." It's "just what Elon Musk wants you to see." If you see stuff he doesn't like he'll step in to make sure you don't see it for long.
Reposted by Ryan J. Gallagher
Factor analysis and varimax have been widely misunderstood among statisticians. Two years ago, Muzhe Zeng and I had a discussion paper at JRSS-B that tried to clarify why factor analysis and varimax were actually decades ahead of their time
It’s free (with discussion) here:
doi.org/10.1093/jrss...
It’s free (with discussion) here:
doi.org/10.1093/jrss...
Vintage factor analysis with Varimax performs statistical inference
Abstract. In the 1930s, Psychologists began developing Multiple-Factor Analysis to decompose multivariate data into a small number of interpretable factors
doi.org
November 20, 2025 at 6:28 PM
Factor analysis and varimax have been widely misunderstood among statisticians. Two years ago, Muzhe Zeng and I had a discussion paper at JRSS-B that tried to clarify why factor analysis and varimax were actually decades ahead of their time
It’s free (with discussion) here:
doi.org/10.1093/jrss...
It’s free (with discussion) here:
doi.org/10.1093/jrss...