Introducing the Concept Grants from Sage Doing Research category winner: Y: The Home of LLMs

by Giulio Rossetti

Unleashing the Power of Social Simulations:
Y Social, an LLM-powered Social Media Digital Twin

The digital age has brought unprecedented opportunities and challenges, particularly in studying social phenomena like echo chambers, polarization, and the ever-evolving effects of AI-driven content. In an era where online platforms increasingly mediate social interactions, understanding social dynamics has never been more critical.

To cope with the rapidly evolving online world, we developed Y - a Social Media platform Digital Twin powered by Large Language Model (LLM) agents.

What is Y?

Y is not just another simulation tool. It is a versatile platform designed to explore "what-if" social scenarios in a controlled environment. Unlike traditional agent-based models, Y leverages the power of LLMs to simulate more nuanced and unpredictable social behaviors.

The strength of Y lies in its ability to create realistic, synthetic data (i.e., social interactions and user-generated content) mirroring real-world social media platforms.

But Y goes beyond data generation. It enables designing and testing interventions to counter or enhance specific social dynamics. For example, Y could be used for simulating the impact of a new algorithm designed to reduce polarization or to study to what extent/if LLM interactions can capture known social patterns.

Under the hood: Personalization and Flexibility

Y's agents are highly customizable, allowing for the creation of heterogeneous synthetic populations reflecting the complex facets of real-world users. Agents’ behavior can be refined by specifying characteristics like age, political leaning, personality traits, education level, interests, and activity levels - or even the recommender system they interact with.
Agents can access online news sources, decide who to follow/unfollow, and determine what content to produce, share, comment on, or react to, thus ensuring that simulations are as close to real-world scenarios as possible.

A Tool for the Research Community

What sets Y apart from existing tools is its free software nature. While most Digital Twins are owned and operated by big-tech industries, Y is being developed as a free software alternative. We believe that by making it accessible to a broad, multidisciplinary community, we can foster innovation and collaboration in ways that proprietary tools cannot.

Whether the goal is to conduct social network analysis, content analysis, or explore the cognitive aspects of LLMs, Y represents a viable solution to test your hypotheses and finetune your methodology.

The Road Ahead

We plan to use the SAGE grant funding to finalize the first easy-to-use platform version. Our primary goal will be to develop a user-friendly web interface allowing non-technical users to easily define, configure, and run simulations. We want Y to be as accessible as possible, enabling users with heterogeneous backgrounds to leverage its capabilities without extensive technical knowledge.

Conclusion

Y represents a significant leap forward in social simulation, offering a powerful, flexible, and open-source tool to study online social phenomena. We are excited about the potential of Y to transform social science research, and we look forward to seeing the innovative ways in which the research community will use this tool.

Y is a tool for the community: explore the platform, experiment with its features, and let us know what you think—your feedback is crucial to making it the best it can be.

Project website: https://ysocialtwin.github.io/
GitHub: https://github.com/YSocialTwin
Preprint: https://www.arxiv.org/abs/2408.00818