Dave's latest is smart and worth a read (even without the kind words).
But I'd phrase it differently. It's not that that era of optimization ended, per se—it's that WHAT was optimized, and for whom, changed.
Exhibit A: www.nytimes.com/2025/12/09/b...
There's a colorable argument that a substantial part of the effectiveness of most of the major innovations in campaign communication in the last 15 years have been as a result of novel strategies to gain attention, including but not limited to social pressure GOTV mail, relational organizing, etc
🇺🇸 #July4 project 🇺🇸
How does ChatGPT map the Declaration’s 27 #grievances to 2025 executive actions?
Scale 0-1 minor · 2 noticeable · 3 substantial · 4 ≈ average 1776 abuse · 5 worse.
AI coded (o3). Thread follows ⤵︎
[1/66]
Academia in 2025: For weeks I’ve been working through botched copy-edits of my monograph, outsourced by the publisher to a company I’d never heard of, until it finally dawned on me the terrible job might be AI. A quick search confirmed the company recently launched new AI software, which now means..
Just published: our new paper in Scientific Reports
📄 “Coordinated Link Sharing on Facebook”
doi.org/10.1038/s415...
We introduce a statistically grounded, human-interpretable method to detect coordination on social media.
ATTENTION: NSF GRANT RECIPIENTS
We received a heads up from a trusted source that you should proactively download/print/screen shot any documentation on research.gov pertaining to your NSF awards, both those that are current and any that have closed in the last 5-6 years.
1/n
Thrilled to have this published.
Everything you've wanted to know about political Youtube -- from Kevin Munger, Jim Bisbee (@jamesbisbee.bsky.social), Omer Yalcin (@ofyalcin.bsky.social), Joe Phillips (@polpsychjoe.bsky.social), and myself.
Out now in the Journal of Quantitative Description.
list of banned keywords
🚨BREAKING. From a program officer at the National Science Foundation, a list of keywords that can cause a grant to be pulled. I will be sharing screenshots of these keywords along with a decision tree. Please share widely. This is a crisis for academic freedom & science.
New and important: we built a federal expenditure tracker. All expenditure line items that are available on the Daily Treasury Statement.
USAID was zeroed out on 1/28 and has been at zero ever since.
www.hamiltonproject.org/data/trackin...
#EconSky
Arguably the biggest national security breach in U.S. history. Private employees downloading personnel data on every federal employee and tax and social security data on every American onto private unsecured servers. Needless to say, completely illegal and subject to major prison sentences.
To say François Chollet has been a skeptic of current LLM capabilities is an understatement.
So this post from him is all the more remarkable.
arcprize.org/blog/oai-o3-...
These are true “holy shit” results — a huge step change in AI capabilities on the very hardest current benchmarks.
For competitive coding, this is like Deep Blue beating Kasparov.
We all need to update our priors on how much knowledge work AI will displace — and how quickly.
The key finding, though, is that right-wing rhetoric on YouTube isn't simply echoing media messages - it's being actively transformed through user interactions.
Understanding online extremism requires studying the role of users themselves in that transformation process.
So where ARE users learning these extreme associations, if not from the original videos?
Our methods emphasize the most-engaged content. Commenters may be selectively remixing content from other platforms or less-seen, more-extreme channels like OANN. But this is an important ? for future work.
For example, videos on BLM protests focused on news events. But commenters added entirely new associations-- "antifa," "marxist," etc.--the videos never mentioned.
There is also a big Trump effect: he becomes the central node in comment networks despite a modest presence in the original coverage.
Our results look much more like the networked framing story.
While media outlets set the broad agenda (COVID, BLM, election), commenters consistently introduce conspiracy theories and emotional rhetoric absent from the original videos.
Networked framing research (e.g. Meraz & @zizip.bsky.social) has argued that user discussion on social media actively transforms frames in news coverage. By contrast, network agenda setting argues that the media transfers bundles of associations more-or-less unchanged.
Which do we find?
Our data includes every video, w/ comments, from the Fox News, OANN, Daily Wire, and Breitbart YouTube channels from 2019-2021.
We use semantic network analysis to track how language changes between the 19,112 video transcripts and the 661,958,464 comments.
Where does extreme right-wing rhetoric come from?
In a new article in Communication Research, Yuan Hsiao and I analyze 19K videos & 661M comments on conservative YouTube channels to trace how extreme rhetoric forms.
BLUF: it comes mostly from users themselves.
doi.org/10.1177/0093...
Scatterplot of term centrality score for the transcripts network (i.e. the original video) and the comments network. Green terms indicate terms with a higher centrality score in the transcripts network. Red terms indicate terms with a higher centrality score in the comments network.
For example, videos on BLM protests focused on news events. But commenters added entirely new associations--"antifa," "marxist," etc.--the videos never mentioned.
There is an enormous Trump effect: Trump becomes the central node in comment networks despite little presence in the original coverage.
Our results look much more like the networked framing story.
While media outlets set the broad agenda (COVID, BLM, election), commenters consistently introduce conspiracy theories and emotional rhetoric absent from the original videos.
Networked framing research (e.g. Meraz & @zizip.bsky.social) has argued that user discussion on social media actively transforms frames in news coverage. By contrast, network agenda setting argues that the media transfers bundles of associations more-or-less unchanged.
Which do we find?
Our data includes every video, w/ comments, from the Fox News, OANN, Daily Wire, and Breitbart YouTube channels from 2019-2021.
We use semantic network analysis to track how language changes between the 19,112 video transcripts and the 661,958,464 comments.
Awesome job alert: Analyst Institute is hiring a research manager. Great job for Social Science PhD who wants to change the world for the better through experiments.
analystinstitute.pinpointhq.com/en/postings/...
Or to put it another way:
bsky.app/profile/jake...
This is a really great piece, as Ganz's work always is. I think the inconvenient thing is the civic associationism of American history was heavily driven by local religious congregations, supplemented in the 20th century by veterans groups like the VFW. It was these groups that included "normies."
Yes—but the most important shift needed here is simpler.
We need less bullshit about “listening,” and more of the hard work of audience building.
A good time to (re)read @newsprof1.bsky.social @melbunce.bsky.social @martinscott2010.bsky.social on government capture of public media and its relationship to democratic backsliding global.oup.com/academic/pro...
The AI military-industrial complex is forming quickly—legacy contractors, big tech firms, and generative AI startups. I wrote about what the military and the AI industry want for @gzeromedia.bsky.social:
This looks to be the story Penny's team referenced. It's an AI-written article by "The Pinnacle Gazette," a site run by a popular Turkish science company.
The jury is anonymous, but "Martin Beck" is not one of their names, according to Penny's lawyers.
AI in journalism, folks.