Sign in

Ryan J. Gallagher

@ryanjgallag.com
4.7K followers 1.1K following 568 posts

Applied scientist trying to make the internet a little better. PhD. Trust & safety, networks, fingerstyle guitar. I use my hair to express myself. They/he

PostsRepliesMedia
Reposted by Ryan J. Gallagher
Marco @mcognetta.bsky.social · 7h
🚨 [Token][ization] Paper Alert 🚨 Tokenization is a wildly understudied area of language modeling despite it having effects across all of NLP. Over the past ~8 months, 32 (!) tokenizer researchers put together the most comprehensive survey of the field. Check it out!
19628
Reposted by Ryan J. Gallagher
Alexios Mantzarlis @mantzarlis.com · 10h
Today on @indicator.media: Google is breaking its promise to label ads for unofficial government services providers. Travelers are paying the price.
indicator.media
Google promised to label ads for unofficial visa providers. It’s doing a lousy job of it.
An Indicator audit found labels on only 8% of ads for visa providers that overcharge their users
165
Reposted by Ryan J. Gallagher
Maria Antoniak @mariaa.bsky.social · 29/09/2026
I'm recruiting 1-2 PhD students to join our lab in Fall 2027, through either Computer Science or Information Science at the University of Colorado Boulder! Looking for people with interests in NLP plus [healthcare | literary studies | narratives | social media | etc.]. Join us! 🏔️☀️
cls-lab.com
CLS Lab — Culture, Language, & Systems | University of Colorado Boulder
The Culture, Language, & Systems Lab at CU Boulder studies the language systems that transmit and shape modern culture, from internet platforms to literary archives to language models.
06950
Reposted by Ryan J. Gallagher
AT Protocol Developers @atproto.com · 28/09/2026
The organization that governs the PLC Directory now formally exists as a registered Swiss Association and has taken the first steps towards being able to independently operate the directory. blog.plcred.org/3mwlphq42d227
blog.plcred.org
First steps of the PLC organization
213432
Ryan J. Gallagher @ryanjgallag.com · 26/09/2026
I would be insufferable if I ever taught Python again
Four panel comic. First label is a bird labeled "New programmers" rejecting a cracker called "Explicit type declarations." It takes a bite and over the panels the bird realizes it loves the cracker
030
Reposted by Ryan J. Gallagher
Tomás G. @tomasgna.bsky.social · 24/09/2026
i have a new paper out in Social Media+Society about Trust & Safety specialists! 📝 i argue that they construct online harms by presenting themselves as advocates for users, creating risks, and publicizing this work. my goal was to make sense of the inherent instability of T&S work...
162
Reposted by Ryan J. Gallagher
Rude1 Haunted Badness. ⁂ @rude1.blacksky.team · 23/09/2026
tawk. 🗣️
Rudy mozilla wildposting
24463128
Reposted by Ryan J. Gallagher
Joe Bak-Coleman @jbakcoleman.bsky.social · 20/09/2026
We can simulate fish schools or ants in ways that solve complex problems (gradient detection, bridge building, nest selection) but no one would create this weird soul/no-soul dichotomy to argue they these that multi-agent swarms of extremely simple rules are “thinking”.
14504169
Reposted by Ryan J. Gallagher
Joe Bak-Coleman @jbakcoleman.bsky.social · 20/09/2026
For whatever reason the fact that LLMs manifest their task completion with language has had folks wanting to jump the gun on a big thorny question of whether/when they “think” in the sense that we all have come to understand it. They may simply not need “think” to to complete complex tasks.
723231
Ryan J. Gallagher @ryanjgallag.com · 19/09/2026
Thank you for this. Several parts are what I've been thinking verbatim as I navigate this new way of engineering both personally and professionally. I appreciate knowing other people feel the same way, and are working through similar tensions
010
Reposted by Ryan J. Gallagher
ewan @ewancroft.uk · 19/09/2026
AI is a tool. Not a replacement.
blog.ewancroft.uk
I Still Built It
if an LLM wrote a significant amount of the implementation, apparently the human directing it no longer counts.
3365
Reposted by Ryan J. Gallagher
Alvin Zhou @alvinyxz.bsky.social · 18/09/2026
🎉 New editorial: "Be Careful What You Prompt For: Generative AI in Computational Communication Research." It opens our special issue of 12 open-access papers, co-edited with @wrahool.bsky.social and @ewam.bsky.social 🧵 doi.org/10.5117/CCR2...
doi.org
Be Careful What You Prompt For: Generative AI in Computational Communication Research | Amsterdam University Press Journals Online
Abstract Generative artificial intelligence (GenAI) has field-level implications for computational communication research (CCR), not only by expanding methodological repertoires, but also by reshaping...
199
Reposted by Ryan J. Gallagher
Carl T. Bergstrom @carlbergstrom.com · 19/09/2026
When we talk about the costs that LLMs impose on society we should not forget the fact that nothing works anymore because everyone is so scared of getting their shit scraped. I can't even use google scholar from my university or starlink because they send too much traffic.
Google	
Sorry...
We're sorry...

... but your computer or network may be sending automated queries. To protect our users, we can't process your request right now.
See Google Help for more information.
984454963
Reposted by Ryan J. Gallagher
Chris Hayes @chrislhayes.bsky.social · 18/09/2026
the experience of COVID really drove this home for me. It wans't the apocalypse, life went on, but a million people died and there were enormous social costs, some of which continue to this day and then when it was done most people were like "let's never think about that again"
792365392
Ryan J. Gallagher @ryanjgallag.com · 18/09/2026
me, every time I start a new job, annoying every single coworker: have you heard about our lord and savior pytest
000
Reposted by Ryan J. Gallagher
Per Engzell @pengzell.bsky.social · 18/09/2026
A paper is finished when the embarrassment of submitting it becomes smaller than the embarrassment of still working on it
630254
Reposted by Ryan J. Gallagher
Dave Willner @dwillner.bsky.social · 16/09/2026
At TrustCon this year I talked about a technique we’ve developed for automatically optimizing content-moderation policies, using an inversion of the binocular labeling approach Zentropi had already pioneered. Today we're shipping the tool that technique became. blog.zentropi.ai/optimizing-o...
blog.zentropi.ai
Optimizing our Policy Optimizers
Today, we are releasing our next-generation policy refinement tools: policy-only correction, label-only correction, and auto-optimization.
3176
Reposted by Ryan J. Gallagher
Vladimir Salnikov @v4ldelund.bsky.social · 16/09/2026
"just look at the data" final boss
316022
Ryan J. Gallagher @ryanjgallag.com · 16/09/2026
"production" for me is just missing some predictions for a half hour before rolling back, at least I can't take down anything user facing
010
Ryan J. Gallagher @ryanjgallag.com · 16/09/2026
Everything is fine, and I know this is the usual rite of passage, but Feels Bad
130
Ryan J. Gallagher @ryanjgallag.com · 16/09/2026
Caused my first production incident at the new job 👏🏻😭
170
Reposted by Ryan J. Gallagher
Conspirador Norteño @conspirator0.bsky.social · 15/09/2026
For over a year, an unknown entity has been hijacking Bluesky accounts and incorporating them into a spam network that follows real users while serving up a mix of political posts, news links, and plagiarized photos. In recent months, some of the spam accounts' posts have started to go viral.
collage of the profiles of 24 Bluesky spam accounts that engage in bulk follow activity
11396223
Reposted by Ryan J. Gallagher
Micah @rincewind.run · 14/09/2026
I do not think anyone has to leave twitter for bluesky but I do think everyone has to leave twitter
223225579
Reposted by Ryan J. Gallagher
Amy Zhang @axz.bsky.social · 14/09/2026
We were interested in studying the Bsky custom feed ecosystem, with a focus on feed creators, as arguably the first real instantiation of the vision of middleware providers for social media services like recommendation. The idea always seemed great in theory, but how sustainable is it really?
34315
Reposted by Ryan J. Gallagher
Jacky Alciné @jacky.wtf · 23/08/2026
If you're moderately technical (aka if you have the inklings of understanding of a programming language and/or know of TCP/IP); this course from @blackskyweb.xyz might be for you if you want to understand how this stuff works (like how does Blacksky/Bluesky work) learn.blacksky.community
A list of modules explaining how ATProto works
36421
Reposted by Ryan J. Gallagher
Information, Communication & Society @icsjournal.bsky.social · 14/09/2026
#OutNow in #iCS Research on coordinated social media manipulation has grown rapidly, but the field lacks a systematic synthesis of empirical findings on observed campaigns. This study addresses that gap through a systematic review of 83 studies. www.tandfonline.com/doi/full/10....
tandfonline.com
Just the tip of the iceberg? State of the art of coordinated social media manipulation research
Social media environments are increasingly exploited by manipulative actors through coordinated social media manipulation (CSMM) campaigns: the intentional and deceptive orchestration of social med...
011
Reposted by Ryan J. Gallagher
danah boyd @zephoria.bsky.social · 14/09/2026
PhD students (& new PhDs): I'm hiring a postdoc at Cornell (Ithaca) to conduct a novel study at the intersection of political economy and tech. Applications are due Oct 16. There are a LOT more details in the job ad so make sure to read it thoroughly: academicjobsonline.org/ajo/jobs/32502
lnkd.in
LinkedIn
This link will take you to a page that’s not on LinkedIn
22221
Reposted by Ryan J. Gallagher
Brandy Zadrozny @brandyzadrozny.bsky.social · 11/09/2026
Get in, folks: a new Russian disinfo campaign is targeting the midterms, specifically Democrats, in what seems to be the first attempt by the Kremlin-backed op to meddle in this year’s U.S. elections. They're faking celebrity videos attacking Dems and are...very stupid. www.ms.now/news/russia-...
ms.now
A Russian disinformation campaign is doctoring celebrity videos to meddle in the midterms
The Kremlin-backed operation, known as Matryoshka, has Hollywood actors telling voters to disavow the Democratic Party and vote Republican.
8824581549
Reposted by Ryan J. Gallagher
Brandy Zadrozny @brandyzadrozny.bsky.social · 11/09/2026
This story is also a tale of two social media platforms. Seven of the videos were posted to Bluesky, but most were posted to X. Bluesky removed the inauthentic accounts. X did nothing.
5415125
Reposted by Ryan J. Gallagher
kate conger @kateconger.com · 11/09/2026
Over six months, we tracked the appearance of child sexual abuse material on X. We found images from the National Center for Missing and Exploited Children's database, which is considered one of the most highly vetted and which many tech companies block on upload. www.nytimes.com/2026/09/11/t...
nytimes.com
Elon Musk Has Pledged to Rid His Platform, X, of Child Sexual Abuse, but It Persists
Reviews by the Canadian Center for Child Protection and The New York Times found that explicit images of children remain on the social media site owned by Elon Musk.
122068929
Ryan J. Gallagher @ryanjgallag.com · 11/09/2026
the people yearn for open networks even if they don't know it. think of how much content is screenshots across platforms, and think of how it could just be all on the same protocol so you can share it anywhere
040
Reposted by Ryan J. Gallagher
Maria Antoniak @mariaa.bsky.social · 10/09/2026
New from our lab! #COLM2026 When people generate stories, they don't just write one prompt. Instead, they explore narrative space via branching edits 🌱 We reconstruct 24k of these edit trees 🌳 from chat logs and map edit types, story formats, how they relate to tree depth, and more!
The Garden of Forking Prompts: How Users Explore Narrative
Space in Story Generation
Advait Deshmukh♣ Nora Benedict♠ Melanie Walsh♡ Maria Antoniak♣
♣University of Colorado Boulder ♠University of Georgia ♡University of Washington

Abstract

Large language models (LLMs) have changed the way people engage
with stories. Drawing on public chatbot logs, we can see that when users
generate stories, they iteratively edit their prompts to explore narrative
possibilities, adjusting characters, redirecting plots, and swapping fictional universes. As aggregated data, these prompts represent rich traces of creative preference at scale. Yet story generation evaluation benchmarks rely on static, one-shot prompts that cannot capture this exploratory behavior. In this work, we study how users revise consecutive story prompts in the wild. Using a dataset of naturally occurring user-chatbot conversations, we construct WildStories, a sample of 275,635 story generation prompts (labeled with story format, prompt components, and explicitness), and WildEdits, a collection of 24,291 edit trees that model how users iteratively edit base story prompts and explore branching story possibilities. From these trees we develop a framework of edit types crossing four directions (adding, removing, changing, and extending) with fourteen targets (e.g., plot, character, genre). We then use our datasets and this framework to
analyze user behavior in navigating narrative space via LLMs. Finally, we
show how automated permutations based on the framework can be used
for story generation benchmarking. Content Warning: This paper works with “wild” chatbot logs, which often include toxic and sexually explicit themes.
29629
Reposted by Ryan J. Gallagher
Alexios Mantzarlis @mantzarlis.com · 10/09/2026
New on @indicator.media: Community Notes on Instagram covered just 3% of false posts debunked by the US bureau of AFP Fact Check. This analysis comes one day after Meta announced it would expand the program to 16 new countries.
indicator.media
Meta says Community Notes is bigger than fact-checking. That's not the whole story
Crowdsourced fact-checking on Instagram covered just 3% of false posts debunked by one fact-checking website
1169
Ryan J. Gallagher @ryanjgallag.com · 10/09/2026
Who is going to be the big powerful player they eventually try to hack that says, "Yeah, actually no, I'll see you in court"
020
Reposted by Ryan J. Gallagher
Colin @colin-fraser.net · 10/09/2026
I'm having some fun in this thread but I do think you should get in trouble for unleashing insane hacker robots on the open Internet that you've explicitly instructed to attack critical public infrastructure, which, to be clear, it eventually succeeds at.
745074
Ryan J. Gallagher @ryanjgallag.com · 10/09/2026
I mean I know why, just how did we get here
120
Ryan J. Gallagher @ryanjgallag.com · 10/09/2026
I don't really understand why we're letting AI companies run "tests" that involve real hacking and supply chain attacks Can you imagine if they tried to publish that ten years ago? It would have been a bigger scandal than the FB emotional contagion study and ended multiple careers
181
Reposted by Ryan J. Gallagher
Mike Sager @mikesager.net · 10/09/2026
If you write a math program to predict text fairly accurately and then build a layer to run other programs, and then spec your programs to use your language layer and design a task for it to find cybersecurity vulnerabilities, it is going to do that using vulnerable websites.
14414
Reposted by Ryan J. Gallagher
Joshua Foust 🪖🎮 @joshuafoust.com · 08/09/2026
This sucks for the researchers, but also this is why you should only use an Enterprise license if you use AI for research, since that comes with data protections that would make this behavior very actionable., If you use a private account, however, any institutional data protections don’t apply.
22810
Reposted by Ryan J. Gallagher
Stanford Tech Impact and Policy Center @techimpactpolicy.bsky.social · 08/09/2026
The Journal of Online Trust and Safety is thrilled to release its new special issue: Digital Intersectionality and Marginalization in the Majority World. 🌏 🔗 tsjournal.org/index.php/jo... #JOTS #TrustAndSafety #MajorityWorld #GlobalSouth
155
Reposted by Ryan J. Gallagher
Quinta Jurecic @qjurecic.bsky.social · 08/09/2026
The Navier-Stokes fight is a perfect encapsulation of where AI development is right now: this should be really cool and exciting, and instead because of Silicon Valley egos it's become a fight over cheating, surveillance, and companies trying to get one up over the other
749682
Reposted by Ryan J. Gallagher
Dave Karpf @davekarpf.bsky.social · 08/09/2026
My AI takes: -data centers: still bad. -A.I. for coding: if the coders say it works, I'm not gonna fight them over that. -How big of a deal we make out of A.I. coding really depends on where you sit. -A.I. isn't vaporware... but the A.I. future ABSOLUTELY is. davekarpf.beehiiv.com/p/a-few-note...
davekarpf.beehiiv.com
A few notes on the state of AI right now
A.I. isn't vaporware. But the A.I. future ABSOLUTELY is.
1218332
Ryan J. Gallagher @ryanjgallag.com · 06/09/2026
What books are best to read more about the (slave) labor conditions of people coerced into running online scams?
275
Reposted by Ryan J. Gallagher
ROOST @roost.tools · 04/09/2026
We’re holding our Osprey Working Group call in just under an hour and a half (1730 UTC)! This public meeting brings adopters, contributors, and curious folks together to discuss and plan the Osprey open source project that helps power T&S at Bluesky, Matrix, Discord, & more. Come join us!
github.com
September 4, 2026 · roostorg osprey · Discussion #484
Our bi-weekly working group call is Friday 1730–1830 UTC! Google Meet #osprey in the ROOST Discord Proposed discussion topics: Adopters! Pain points, feedback, questions, etc. ROOST/Community updat...
062
Ryan J. Gallagher @ryanjgallag.com · 04/09/2026
I'm a sucker for these quizzes even though it's obviously the same thing every time
010
Ryan J. Gallagher @ryanjgallag.com · 04/09/2026
Kind of funny I'm geographically right between Portland and Boston (New Hampshire). Not sure where that Miami is coming from
Cities most like my language: Portland ME, Boston, Miami. Least: New Orleans, Houston, DallasWhat gave my language away: rotary, aunt, sneakers, pecan
100
Ryan J. Gallagher @ryanjgallag.com · 04/09/2026
they got me unfortunately
Heat map of the US indicating most similar language in the northeast, least similar in the deep south
110
Reposted by Ryan J. Gallagher
Joshua Foust 🪖🎮 @joshuafoust.com · 04/09/2026
“Communicative strategies of encouraging suspension of disbelief and invoking deictic references to the [present] in order to leverage the authority of the dead... raises questions around the rights and responsibilities of publics, of digital platforms, and of the dead themselves.” #CommSky
doi.org
‘This here is a true representation of who I was’: Synthetic media and the authority of the dead - Graham Meikle, Johanna Sumiala, 2026
The dead are increasingly being resurrected through synthetic media. This article draws inspiration from the histories of both manipulated media and mediations ...
021
Reposted by Ryan J. Gallagher
jon ben-menachem @jbenmenachem.com · 02/09/2026
The slop factory is now claiming to have invented iterative, abductive research design… 20th century ethnography would like a word, sir.
Alfred Wahlforss in • 3rd+
+ Follow
...
CEO & Co-Founder @ Listen Labs...
Book an appointment
5d • G
Anthropic coined a new research term: "self healing studies."
Historically, it was hard to adapt research studies after they launch. The researcher writes the study guide, the interviews run, and if users raise something new along the way, it has to wait for the next study.
In a "self healing study", there is a continuous feedback loop between interviews and the study guide. Each round of interviews identifies new topics, they are added to the guide, and the next interviews explore them.
Nothing waits for a follow-up study - the guide keeps evolving with the study.
And the researcher sits above all of it as the guardrails.
The study proposes additional topics, and the researcher uses their judgment to decide which ones deserve exploring.
The term comes from Jane Justice Leibrock's team at Anthropic, from their churn interviews on Claude Code.
Before we even had an MCP, Anthropic built its own plugin to run Listen studies because they wanted them running around the clock.
That's the future of research.
311643
Reposted by Ryan J. Gallagher
Greta Warren @gretawarren.bsky.social · 02/09/2026
How do people use images to spread misinformation online?🖼️ We developed a taxonomy, analysed 27k posts from X in 7 languages & found: 🔹Slanted framing of real images is more common than deepfakes and doctored images 🔹Vaccine misinfo borrows credibility via news screenshots arxiv.org/abs/2608.29681
054