6 papers
Do Modules Stay in Their Lane? Role Drift in Compound LLM Systems
Xiaoyang Cao, Siddarth Srinivasan, Michiel A. Bakker
End-to-end reinforcement learning can improve the accuracy of compound LLM systems, but it does not constrain how modules divide labor internally. We identify Role Drift, a failure…
AI Organizations are More Effective but Less Aligned than Individual Agents
Judy Hanwen Shen, Daniel Zhu, Siddarth Srinivasan +5
AI is increasingly deployed in multi-agent systems; however, most research considers only the behavior of individual models. We experimentally show that multi-agent "AI organizatio…
The Prosocial Ranking Challenge: Reducing Polarization on Social Media without Sacrificing Engagement
Jonathan Stray, Ian Baker, George Beknazar-Yuzbashev +42
We report the first direct comparisons of multiple alternative social media algorithms on multiple platforms on outcomes of societal interest. We used a browser extension to modify…
WOMAC: A Mechanism For Prediction Competitions
Siddarth Srinivasan, Tao Lin, Connacher Murphy +3
Competitions are widely used to identify top performers in judgmental forecasting and machine learning, and the standard competition design ranks competitors based on their cumulat…
Tell Me Why: Incentivizing Explanations
Siddarth Srinivasan, Ezra Karger, Michiel Bakker +1
Common sense suggests that when individuals explain why they believe something, we can arrive at more accurate conclusions than when they simply state what they believe. Yet, there…
Self-Resolving Prediction Markets for Unverifiable Outcomes
Siddarth Srinivasan, Ezra Karger, Yiling Chen
Prediction markets elicit and aggregate beliefs by paying agents based on how close their predictions are to a verifiable future outcome. However, outcomes of many important questi…