LLMs and Online Forums: A Framework for Sustainable Collaboration
A new research paper proposes a framework for sequential interaction between generative AI systems and online Q&A forums, addressing the paradox that while LLMs draw users away from these forums, they depend on forum data for improvement. The framework models non-monetary exchanges, asymmetric information, and incentive misalignment. Using real Stack Exchange data and common LLMs, simulations show that despite incentive misalignment, players can achieve roughly half the utility of an ideal full-information scenario, highlighting potential for sustainable collaboration.
Key facts
- Generative AI systems draw users away from Q&A forums but depend on forum data.
- Framework proposes sequential interaction where AI proposes questions and forum publishes some.
- Framework captures non-monetary exchanges, asymmetric information, and incentive misalignment.
- Simulations use real Stack Exchange data and commonly used LLMs.
- Incentive misalignment is demonstrated empirically.
- Players can achieve roughly half the utility of an ideal full-information scenario.
- Results highlight potential for sustainable collaboration.
- Paper is from arXiv, Computer Science > Artificial Intelligence.
Entities
Institutions
- arXiv
- Stack Exchange