Big shoutout to my advisors: @haewoon.bsky.social @jisunan.bsky.social for their guidance and support on this project!

📄 Paper link: arxiv.org/pdf/2510.07488

🔮 As LLMs evolve from individual tools to collaborative agents, understanding their team dynamics becomes essential.

Applying team science offers a powerful lens for designing AI systems that collaborate effectively, leverage diversity, and adapt like human teams do.

🧠 Key findings
🧩 Agents tend to overestimate team performance before the task; post-task reflections surface misalignment and integration difficulties.

🧑‍⚖️ Evaluations via GPT-4o (“LLM-as-a-judge”) agree that flat teams score better in comprehension, coherence, reasoning, and confidence.

🧠 Key findings

⚖️ Flat (decentralized) teams often outperform hierarchical ones on reasoning tasks.

⚔️ Diversity is a double-edged sword: it can enrich reasoning in some settings, but also introduce coordination friction especially under hierarchical structure.

We designed flat and hierarchical teams of LLM agents, assigning them personas (e.g. demographics) to inject controlled diversity. These teams were tested on reasoning and social reasoning tasks. We combined quantitative performance evaluation + qualitative analysis to analyze interaction dynamics

New preprint! 🚀 We ask: What happens when you bring human team science into the design of multi-agent LLM systems? In particular, how do team structure, diversity, and interaction dynamics influence how AI agents collaborate?
📄Full Paper: arxiv.org/pdf/2510.07488
👩‍💻Code: github.com/Rasikamurali...

New jobs for computational social scientists at UNC and USC: unc.peopleadmin.com/postings/307... amd uscjobs.sc.edu/postings/194...

🚨 First OSoMe Awesome Speaker of the year! 🚨

📅 Sept 12 | 12pm ET
🎤 Patrick Warren (Clemson University)
📍 Zoom

Measuring the Impact of a Large State-Sponsored Narrative-Laundering Campaign: The case of the Storm-1516 attack on Zelensky

🔗 Register: iu.zoom.us/meeting/regi...

In Norrköping, Sweden for #IC2S2!
Excited to present my collaborative work with @baottruong.bsky.social and @yyahn.bsky.social!

🔎 Where to find us:
Scaling of Community Rules Across Mastodon Servers
🚨 Parallel Talk- Social Media & Networks| Jul 24, 11 AM | Vingen 3&4

So about ten years ago I found a small network of bots on Reddit dedicated to spreading hate. The thing about it is that these weren't even responding to political topics - they were just responding to random keywords with insults and hate.

The same bots are now here. You should understand why. 🧵