We’d love to see others explore and build on this dataset for research on personality structure, impression representation, cultural evolution, and beyond. Don't hesitate to give us your feedback!
For anyone who didn’t get a chance to click during the session (or wants to revisit), here’s the interactive online interface I mentioned, where you can explore and use the data: societyinmind.net:5500/assets/trait...
It was also my first time presenting my own work at SPSP! The Data Blitz format moves fast, so thank you to everyone who stopped by and engaged. And if anything flew by too quickly, I’m very happy to follow up.
Today at SPSP was busy in the best way — a full, energizing day packed with fascinating talks and posters. I came away with lots of new ideas (and an even longer reading list).
We’d love to see others explore and build on this dataset for research on personality structure, impression representation, cultural evolution, and beyond. Don't hesitate to give us your feedback!
For anyone who didn’t get a chance to click during the session (or wants to revisit), here’s the interactive online interface I mentioned, where you can explore and use the data: societyinmind.net:5500/assets/trait...
It was also my first time presenting my own work at SPSP! The Data Blitz format moves fast, so thank you to everyone who stopped by and engaged. And if anything flew by too quickly, I’m very happy to follow up.
For folks interested in learning about our lab's research, check out this flier with all our presentations at this coming #SPSP2026 conference @spspnews.bsky.social. With research by several rising stars covering tech, culture, politics and more
Credit to our talented lab manager Hanying Yao!
We argue that these problems are often less about insurmountable technical limitations than about incentives and institutional design. From micro-level choices in training to macro-level governance, we lay out plausible avenues for better managing how large AI models prioritize information
2️⃣ Heterogeneity risk: When optimization favors “most likely” outputs under imbalanced signals, diversity in viewpoints and cultural expression can be compressed away—contributing to homogenization, polarization, and stagnation in innovation.
We highlight two downstream risks:
1️⃣ Accuracy risk: When objectives reward fluency, confidence, or engagement more than truth and honesty, errors can become systematic rather than occasional glitches—showing up as hallucinations, sycophancy, and engagement-driven distortions of social perception.
This creates a prioritization problem: when a system must compress, it must decide what to preserve or to discard. This prioritization shapes what people see, learn, and infer about what is true and common. Yet the prioritization of today’s large AI models are often misaligned across multiple ends.
Echoing information-theoretic perspective, learning can be understood as a form of information compression: it exploits regularities in the input to form more efficient representations. And because real-world learning operates under tight capacity constraints, that compression is often lossy.
In this short piece, we argue that many debates about large AI models—both LLMs and recommender systems—become clearer when we treat them as learning systems that must compress information into usable outputs.
Before the end of this year, I’m glad to share a short perspective/policy piece, recently out with @joshcjackson.bsky.social , Zhao Wang, and @williambrady.bsky.social: “Large AI Models Have a Prioritization Problem: Policy Implications and Solutions.”
It’s grad school application season, and I wanted to give some public advice.
Caveats:
-*-*-*-*
> These are my opinions, based on my experiences, they are not secret tricks or guarantees
> They are general guidelines, not meant to cover a host of idiosyncrasies and special cases
In a new paper, we find that sycophantic #AI chatbots make people more extreme--operating like an echo chamber
Yet, people prefer sycophantic chatbots and see them as less biased
Only open-minded people prefer disagreeable chatbots: osf.io/preprints/ps...
Led by @steverathje.bsky.social
Abstract and results summary
🚨 New preprint 🚨
Across 3 experiments (n = 3,285), we found that interacting with sycophantic (or overly agreeable) AI chatbots entrenched attitudes and led to inflated self-perceptions.
Yet, people preferred sycophantic chatbots and viewed them as unbiased!
osf.io/preprints/ps...
Thread 🧵
English language is filled with trait words like “caring” and “smart”
These words are the currency of personality/social psych, yet key questions remain about their evolution, function, and structure
We take on these questions in a preprint led by @yuanzeliu.bsky.social
osf.io/preprints/ps...
Today (w/ @ox.ac.uk @stanford @MIT @LSE) we’re sharing the results of the largest AI persuasion experiments to date: 76k participants, 19 LLMs, 707 political issues.
We examine “levers” of AI persuasion: model scale, post-training, prompting, personalization, & more!
🧵:
On the Malleability of Democratic Attitudes: Do Citizens' Views of Democracy Vary With How They Feel?: https://osf.io/hz2mt
✨Did markets make Americans more cooperative❓🔍
✅YES‼️
Between 1850 and 1920, the US became the largest and most integrated economy in the world 📶🌎
We show that this shift didn’t just move goods and affect prices—it fundamentally changed culture and behavior
🧵 👇 1/17
Great new paper by Stagnaro & Amsalem in NatComm:
Detailed fact-based informational module about gun control leads to persistent reductions in polarization (attitudes move towards midpoint). Seeing the evidence cited by other side makes opinions more moderate www.nature.com/articles/s41...
This paper is now published. www.nature.com/articles/s44...
Treemap chart showing the fragmented landscape of psychological measures.
Want to make nice graphs with me, starting this summer? I am hiring for two PhD positions at the University of Witten/Herdecke.
New preprint on prejudice and state centralization: osf.io/preprints/ps...
Our team of historians, psychologists, and anthropologists analyzed 90 historical societies and Chinese records from 206 BCE - 1911 CE
In both studies, we find a link btw group prejudice and historical state centralization
We are deeply saddened to share the passing of James Liu, Co-Editor-in-Chief of Political Psychology and long-time ISPP member. We are forever grateful for his contributions to our field. Our thoughts are with his loved ones. He will be greatly missed.
ispp.org/news/in-memo...
It has been a challenge to identify predictors of moralization. This new study finds that perceived polarization prospectively predicts attitude moralization during the 2020 US Presidential election
psycnet.apa.org/record/2024-...
#psychology #socialpsyc #PsychSciSky #BehSci Polisky
