Sign in

Joshua Tucker

@jatucker.bsky.social
12.6K followers 1K following 169 posts

Politics Prof, NYU; Director @NYUJordanCenter for Adv Study of #Russia; Co-Director NYU Center for Social Media and Politics @CSMaP_NYU (csmapnyu.org); long suffering NY Mets fan; @j_a_tucker on other platforms; he/him

PostsRepliesMedia
Reposted by Joshua Tucker
Gwyneth McClendon @gmcclendon.bsky.social · 20/09/2026
NYU is hiring in American Politics! Deadline October 6. See more details here: apply.interfolio.com/193724 @rozenas.bsky.social @cdsamii.bsky.social @sanfordgordon.bsky.social @bryantjmoy.bsky.social @awilke.bsky.social @stasavage.bsky.social @jatucker.bsky.social @peterrosendorff.bsky.social
apply.interfolio.com
Apply - Interfolio {{$ctrl.$state.data.pageTitle}} - Apply - Interfolio
02024
Joshua Tucker @jatucker.bsky.social · 19/09/2026
Thanks @tompepinsky.com. I think the finding about potential offline social isolation combined with online friends is particularly important. lnkd.in/p/gUJAxsP9
lnkd.in
White Nationalism Support Found in 5% of US Adults, Higher Among Young White Men | Joshua Tucker posted on the topic | LinkedIn
White nationalism has become more visible in American life, especially in online spaces where young white men spend their time. In a new publication at Nature with Patrick Egan & Jim Bisbee, we examin...
1112
Reposted by Joshua Tucker
Dan Silverman @dmsilverman.bsky.social · 17/09/2026
It’s young (white) men. New study in @nature.com by @jatucker.bsky.social et al looks at the predictors of white nationalist attitudes across an array of survey data. 🔑 Findings: —5% of white adults hold such views —13.5% of white men age 18-29 —Political ideology matters but it’s not just that
215464
Joshua Tucker @jatucker.bsky.social · 14/08/2026
Another related - and much more extensive - version of this that will compare Google w/ and w/o AI summaries and a number of different chatbots has been through multiple pilots and we're about to start data collection shortly as soon as we finalize the PAP.
010
Joshua Tucker @jatucker.bsky.social · 14/08/2026
Working on it! Have a draft we are posting any now that is a follow up to our 2024 Nature article (www.nature.com/articles/s41...) that adds a treatment arm to send people to a chat bot (perplexity) -- will post link here as soon as I have it.
130
Reposted by Joshua Tucker
Brendan Nyhan @brendannyhan.bsky.social · 29/07/2026
🚨New Science Advances🚨 Untrustworthy sources on Facebook and Instagram in 2020: Concentrated exposure but no attitudinal effects www.science.org/doi/10.1126/... -Exposure very low but highly concentrated -Reducing exposure by ~70% for months had no measurable effects *even among frequent consumers*
39637
Joshua Tucker @jatucker.bsky.social · 23/07/2026
Yes, you can apply! Please do!
000
Joshua Tucker @jatucker.bsky.social · 23/07/2026
To apply for the position: csmapnyu.org/jobs and then apply.interfolio.com/181820 Please help us by sharing this announcement widely! 3/3
csmapnyu.org
Work With Us - NYU’s Center for Social Media, AI, and Politics
130
Joshua Tucker @jatucker.bsky.social · 23/07/2026
We will begin reviewing applications on or about *August 1st* and will continue until the position is filled. Virtual job talks/visits to start taking place (hopefully) the following week. For more about the lab, see our website: csmapnyu.org 2/3
csmapnyu.org
NYU's Center for Social Media, AI, and Politics
Strengthening democracy by conducting rigorous research, advancing evidence-based public policy, and training the next generation of scholars.
141
Joshua Tucker @jatucker.bsky.social · 23/07/2026
We have re-opened our search for a 3-year AI-research postdoc at NYU's Center for Social Media, AI, and Politics!! The position starts this **September 1st** (or as soon as possible thereafter) and is in person at NYU. If necessary, we will also consider a 1/1/27 start. 1/3
32934
Reposted by Joshua Tucker
NYU Center for Data Science @nyudatascience.bsky.social · 15/06/2026
"Institutional influence": how governments shape AI chatbots through the training data they ingest — with measurable effects on what the models actually say. A new Nature study by CDS-affiliated Prof. @jatucker.bsky.social and CSMAP's @solmg.bsky.social. 🧵 www.nature.com/articles/s41...
nature.com
State media control influences large language models - Nature
Government-controlled media influences the output of large language models via their training data, and models queried in the languages of countries with lower media freedom show a stronger ...
1118
Reposted by Joshua Tucker
Alex Dukalskis @alexdukalskis.bsky.social · 04/06/2026
“State Media Control Influences Large Language Models.” by @hwaight.bsky.social @eddieyang.bsky.social @reasonyy.bsky.social Solomon Messing, Margaret E. Roberts, @bstewart.bsky.social and @jatucker.bsky.social in @nature.com (2026) #SummerReading
nature.com
State media control influences large language models - Nature
Government-controlled media influences the output of large language models via their training data, and models queried in the languages of countries with lower media freedom show a stronger ...
062
Joshua Tucker @jatucker.bsky.social · 14/05/2026
This paper benefitted enormously from the very involved and insightful editorial team @nature.com, led by @meharpist.bsky.social.
081
Joshua Tucker @jatucker.bsky.social · 13/05/2026
Thanks so much @jenjennings.bsky.social! bsky.app/profile/jenj...
020
Joshua Tucker @jatucker.bsky.social · 13/05/2026
23/ A final takeaway – if social and political institutions help to determine the context in which training data is produced, which in turn affects political leans of LLM outputs, then social scientists have a very important role to play in model development going forward.
152
Joshua Tucker @jatucker.bsky.social · 13/05/2026
22/ Huge thanks to everyone who has given us feedback on this article! We benefitted enormously from your advice and critiques!
100
Joshua Tucker @jatucker.bsky.social · 13/05/2026
21/ The models used in the paper are not the most recent ones. So we replicated the entire study on the latest models, here: state-media-influence-llm.github.io. The new models memorize a great deal more Chinese state media talking points, and are more pro-China in Chinese. Check it out!
state-media-influence-llm.github.io
State Media Control Influences Large Language Models – State Media & LLMs
100
Joshua Tucker @jatucker.bsky.social · 13/05/2026
20/ We’ve also got a post at @goodauth.bsky.social about the article here: goodauthority.org/news/state-m...
goodauthority.org
State media control impacts the output of U.S.-based LLMs
Training data for LLMs does not just fall from the sky, our research finds.
142
Joshua Tucker @jatucker.bsky.social · 13/05/2026
Here’s the actual article @nature.com doi.org/10.1038/s415....
doi.org
State media control influences large language models - Nature
Government-controlled media influences the output of large language models via their training data, and models queried in the languages of countries with lower media freedom show a stronger ...
120
Joshua Tucker @jatucker.bsky.social · 13/05/2026
18/As a final step, we examined these patterns across 37 other countries where at least 70% of the speakers of a language resided in the country. The results were clear: less media freedom = more positive answers in that country’s language about that country’s regime and leaders.
100
Joshua Tucker @jatucker.bsky.social · 13/05/2026
17/ And that finding is robust! Using the WildChat data set, we find similar results when looking at questions asked by actual people. We all got similar results when we used human judges (rather than LLMs).
100
Joshua Tucker @jatucker.bsky.social · 13/05/2026
16/ We found that the answers to questions about Chinese politics produced more favorable answers when asked in Chinese rather than English 70-80% of the time. And, these models are similarly more friendly toward China’s allies like North Korea and Russia.
100
Joshua Tucker @jatucker.bsky.social · 13/05/2026
15/ Since the effect was stronger in Chinese than English, we checked whether prompting in Chinese yielded more favorable answers than prompting in English.
100
Joshua Tucker @jatucker.bsky.social · 13/05/2026
14/ This effect is much stronger in Chinese than English (although interestingly we see non-trivial spillover effects to Japanese)
110
Joshua Tucker @jatucker.bsky.social · 13/05/2026
13/ We found evidence consistent with the idea that pretraining on state coordinated media produced more pro-PRC answers by models:
110
Joshua Tucker @jatucker.bsky.social · 13/05/2026
12/ Having established that this state coordinated media is in the training data, we then ran experiments to see what happened when we added these media sources to an open weight model and continued its pretraining.
120
Joshua Tucker @jatucker.bsky.social · 13/05/2026
11/ But in the newer models (see state-media-influence-llm.github.io) these rates are even higher!
110
Joshua Tucker @jatucker.bsky.social · 13/05/2026
10/ In the paper, we find 3-10% of these phrases are memorized:
110
Joshua Tucker @jatucker.bsky.social · 13/05/2026
9/ But due to opacity around model training, we don’t actually know whether these sources were in the training data. So we look for an “echo” of their presence in training data by prompting LLMs to complete propaganda phrases – which we find.
130
Joshua Tucker @jatucker.bsky.social · 13/05/2026
8/ 3.1M CulturaX documents (1.64%) show substantial phrasing overlap – and this proportion goes up for politically sensitive subjects:
120
Joshua Tucker @jatucker.bsky.social · 13/05/2026
7/ We then check to see the extent to which these media sources end up in CulturaX, a Chinese language part of Common Crawl often used to train frontier LLMs:
130
Joshua Tucker @jatucker.bsky.social · 13/05/2026
6/ We use these articles to understand this influence in models, along with the same from the state media’s digital news application, Xuexi Qiangguo.
130
Joshua Tucker @jatucker.bsky.social · 13/05/2026
5/ We use China as a case study to triangulate how this happens. In other work, co-authors @hwaight.bsky.social , Yin Yuan, @mollyeroberts.bsky.social, and @bstewart.bsky.social found leaked state-scripted talking points reprinted across hundreds of Chinese newspapers: www.pnas.org/doi/10.1073/...
pnas.org
PNAS
Proceedings of the National Academy of Sciences (PNAS), a peer reviewed journal of the National Academy of Sciences (NAS) - an authoritative source of high-impact, original research that broadly spans...
150
Joshua Tucker @jatucker.bsky.social · 13/05/2026
4/Before the findings, this work is an amazing multi-institution collaboration including @csmapnyu.org with co-first authors @hwaight.bsky.social & @eddieyang.bsky.social, plus co-authors: @reasonyy.bsky.social, @solmg.bsky.social, @mollyeroberts.bsky.social rts.bsky.social, & @bstewart.bsky.social
150
Joshua Tucker @jatucker.bsky.social · 13/05/2026
3/ We think this works via control over the media environment & therefore the training data. More state control over the media means less critical LLMs. 6 studies spanning 38 countries, detailed below:
1102
Joshua Tucker @jatucker.bsky.social · 13/05/2026
2/ More specifically, in countries with lower levels of media freedom, when you query models in the language of that country about politics in that country, you are more likely to get a pro-regime/leadership answer than if you ask in a different language.
1167
Joshua Tucker @jatucker.bsky.social · 13/05/2026
1/ Excited to report we have a new paper out @nature.com today! The bottom line: training data for LLMs does not just fall from the sky - it is created in the context of existing social political institutions - and that has consequences for LLM output. nature.com/articles/s41...
nature.com
State media control influences large language models - Nature
Government-controlled media influences the output of large language models via their training data, and models queried in the languages of countries with lower media freedom show a stronger ...
18444
Reposted by Joshua Tucker
Sol Messing @solmg.bsky.social · 13/05/2026
New in Nature: LLMs give "the party line" in the languages of authoritarian regimes. This works when they control the media, which feeds pretraining data. We show more state control over the media means less critical LLMs. 6 studies spanning 38 languages & 13 models. Details ↓
Scatter plot of 38 countries plus China showing the proportion of LLM responses favorable to the regime in the local language vs. World Press Freedom Index score. Lower press freedom predicts more regime-favorable responses; all production models pooled.
7242118
Reposted by Joshua Tucker
Hannah Waight @hwaight.bsky.social · 13/05/2026
I’m excited to share a new paper in Nature that shows how large language models launder the strategic rhetoric of authoritarian states. Paper here: www.nature.com/articles/s41.... A thread.
nature.com
State media control influences large language models - Nature
Government-controlled media influences the output of large language models via their training data, and models queried in the languages of countries with lower media freedom show a stronger ...
211851
Joshua Tucker @jatucker.bsky.social · 12/04/2026
I was using the term as defined by Way and Levitsky, who are very clear that they see Hungary in that category of regime type: The New Competitive Authoritarianism muse.jhu.edu/article/745953
muse.jhu.edu
Project MUSE -- Verification required!
250
Joshua Tucker @jatucker.bsky.social · 12/04/2026
People will be studying Magyar and Tisza's approach for years to come - a lot of valuable lessons here.
051
Reposted by Joshua Tucker
Jen Pan @jenpans.bsky.social · 08/04/2026
New study out in Nature Human Behaviour: 37 million US users were exposed to deceptive networks on Facebook & 3 million on Instagram during the 2020 elections—roughly 15% and 2% of active users. 🧵
1107
Joshua Tucker @jatucker.bsky.social · 12/04/2026
Orban concedes in Hungary. Good reminder that incumbents can lose even in competitive authoritarian regimes if there is enough a wave of popular support for the opposition. www.nytimes.com/2026/04/12/w...
nytimes.com
Hungary’s Orban, Beacon to the Right, Concedes Election Defeat
2378
Reposted by Joshua Tucker
Tiago Ventura @tiagoventura.bsky.social · 01/04/2026
🚨New paper alert🚨 What happens when a polarized democracy bans a social media platform? New work with @cbarrie.bsky.social, Molly Roberts, Chris Schwarz, and @jatucker.bsky.social. We study Brazil's 2024 ban on X and find it created what we define as a "partisan sorting ratchet ." 1/
11910
Reposted by Joshua Tucker
Ken Schultz @kschultz.bsky.social · 04/03/2026
I thought this was an interesting and balanced contribution to the AI discussion here. As an old dog trying to learn the new tricks, let me make a few observations from my own experience using Claude (not Claude code, just the chatbot). 🧵
141
Joshua Tucker @jatucker.bsky.social · 04/03/2026
Thanks!
010
Joshua Tucker @jatucker.bsky.social · 04/03/2026
Thanks @marcratkovic.bsky.social
000
Joshua Tucker @jatucker.bsky.social · 04/03/2026
Probably paid off based on who got most votes in first round regardless of whether it advanced to second round?
010
Joshua Tucker @jatucker.bsky.social · 04/03/2026
Apologies to everyone who has written something interesting on this topic since we sent this in our editors two and a half weeks ago! And huge thanks to our editors at Brookings for their input, excellent editing, and quick turnaround! Full article here: www.brookings.edu/articles/the... 5/5
brookings.edu
The train has left the station: Agentic AI and the future of social science research | Brookings
A new era of agentic AI agents has begun. What does it mean for social scientists? Solomon Messing and Joshua Tucker discuss.
120
Joshua Tucker @jatucker.bsky.social · 04/03/2026
And yes, the title is inspired by Andy Hall's seismic post from January 2nd that "Claude Code and its ilk are coming for the study of politics like a freight train." 4/
110