BDA3 for free statmodeling.stat.columbia.edu/2025/09/13/b...

Chatbot still can’t handle tic-tac-toe statmodeling.stat.columbia.edu/2025/08/24/c...

My goto is to ask LLMs how many states have R in their name. They always fail. GPT 5 included Indiana, Illinois, and Texas in its list. It then asked me if I wanted an alphabetical highlighted map. Sure, why not.

The fact that ChatGPT can produce an essay about the history of blueberries but can't count the letters is a great example of how it is a text producing machine not a reasoning machine.

(real, one-shot :p )

I’m thrilled to share new work out in Social Networks! In this project, randomly allocating students to teams permitted identifying causal effects of co-membership. Thanks to Sarah Gordon, Wayne Baker, Jose Uribe, and Cassandra Chambers for bringing me on! Ful article: 📄 lnkd.in/e9TyvnRX

Do women politicians address violence against women more effectively? It would be a big deal if they do! It would provide (perhaps more) reason(s) to elect more women! But I'm skeptical that this study actually shows that women politicians reduce violence against women. Here's one reason why:

A cartoon by Tom Toro. #NewYorkerCartoons See more cartoons and other funny stuff from New Yorker Humor: nyer.cm/9mf6isU

I understand why people are theorizing that the Grok/South Africa thing was caused by a system prompt, but what it really reminds me of is Golden Gate Claude, which involved artificially amplifying some model features. I am so curious about how this happened. www.anthropic.com/news/golden-...

Gave a talk to our PhD students today, and feeling appreciative of all the recent methods work on improving common practices. Thanks to all the authors who work to make their work accessible, especially on social media. A few articles we discussed:

Talk to me the way Claude talks to me :)

Yemen sparked a digital education revolution with the second phase of “Reading Wikipedia in the Classroom”, which empowered 22 teachers and trainees to boost digital literacy. Led by the Wikimedians of Yemen User Group, volunteers made 1,600+ edits on Arabic Wikipedia. Read more ➡️ w.wiki/_paW2

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.

New publication! Please check out our open access work on a quantitative description of political YouTube—by Kevin Munger, @jamesbisbee.bsky.social, myself, @polpsychjoe.bsky.social, and @matthindman.bsky.social. Thread🧵 by @polpsychjoe.bsky.social below 👇

Our top 10 downloads of 2024. Number 3... How do populist voters differ from other voters and among themselves? Caner Simsek @uni-muenster.de Read more here in 'Voices unheard: How feelings of inefficacy fuel populism' link.springer.com/article/10.1... #Populism #Populistvoting

🎉 My favorite PhD chapter found its home in New Media & Society! I explore whether offline ties matter in online elections, focusing on @wikipedia.bsky.social. In short: They do! Personal connections to voters & candidates affect a user's adminship votes. 🔗 doi.org/10.1177/1461... #PoliSky #SocSky

Hello! It's me.