The Trump assassination attempt *did not* increase support for political violence in the U.S. according to this new paper-- in fact, Republicans expressed *less* support for political violence against Democrats after the event: prlpublic.s3.amazonaws.com/reports/Politi…

🚨New WP: Can LLMs predict results of social science experiments?🚨 Prior work uses LLMs to simulate survey responses, but can they predict results of social science experiments? Across 70 studies, we find striking alignment (r = .85) between simulated and observed effects 🧵👇

I'm on the job market this fall. I study how the state and marginalized populations interact in American politics and policy, with a focus on policy implementation (state capacity) and community organizing and collective action (civic capacity).

Duke is hiring a tenure-track assistant professor of sociology this fall as well as a professor of the practice-- I'd love to see some great young computational social scientists apply! academicjobsonline.org/ajo/jobs/28065 & academicjobsonline.org/ajo/jobs/28071

A group research project incubated at a Summer Institute in Computational Social Science was recently featured in the New York Times! Congrats to all of the co-authors and organizers of our Rutgers location: nytimes.com/2024/07/25/cli…

Can anyone please recommend excellent new work on the ethics of research with LLMs for a new graduate course I'm teaching this fall? Many thanks.

Applications are due September 15th for Duke's Assistant Professor Position in Sociology. Apply here: academicjobsonline.org/ajo/Duke/Socio…

New paper provides a systematic assessment of whether/how LLMs can improve the process of research. TLDR: it's mediocre at most stuff *except* identifying possible ethical issues... interesting finding. pnas.org/doi/10.1073/pn…

Duke is recruiting an Associate or Full Professor for the Rothermere Chair in Technology and Public Policy: academicjobsonline.org/ajo/jobs/28301

Daron Acemoglu has an interesting new paper on the macro-economic impact of AI. His model predicts a modest 0.71% increase in productivity: economics.mit.edu/sites/default/…

😂

GPT-4 reproduces experimental treatment effects in about 75% of 154 studies in top psych journals, but the model is prone to false positives and misestimates confidence intervals: arxiv.org/pdf/2409.00128

Three field experiments that randomized access to Generative AI tools among software developers suggests they enable a 26% increase in task completion: papers.ssrn.com/sol3/papers.cf…

"Knowing that the same content is created by a human expert increases its (reported) perceived quality, but knowing that AI is involved in the creation process does not affect its perceived quality." cambridge.org/core/journals/…

📢 The first chapter of the AI snake oil book by me and Sayash Kapoor is now available online. press.princeton.edu/books/hardcove… It is 30 pages long and summarizes the book’s main arguments. If you start reading now, you won't have to wait long for the rest of the book — it is available to

Judy Woodruff is a national treasure- I was so fortunate to speak to her for this new story, which stands out in my view for its nuanced take on a deeply complex issue: youtube.com/watch?v=3vpfKZ…

What happens if you reward social media users with status badges for expressing open-mindedness? Our Polarization Lab’s new working paper sets out to answer this question via an experiment on our “Social Media Accelerator”: osf.io/preprints/soca….

It was such an honor to welcome Judy Woodruff to the Duke Polarization Lab- unsurprisingly, hers is the most nuanced coverage of this complex issue (social media and polarization) that I've yet seen on television.

New paper finds people prefer ethical guidance generated by GPT-40 over the NYT's "the Ethicist" column. Interestingly, people can discern LLM advice from humans, but this does not diminish their preference: osf.io/preprints/psya…

debunkbot.com, the intervention described in this Science Magazine paper, is now updated to include OpenAI's gpt-o1 (aka strawberry) only a minority of respondents will get gpt-o1, due to costs, but feel free to give it a spin and test your luck!