New project: EvolvingAgents — Bio-Inspired Multi-Agent Evolution 🧬

Excited to kick off EvolvingAgents, a new research project at U Lab!

I’m collaborating with Prof. Gene Robinson and Prof. Jiaxuan You, in collaboration with Prof. James Zou, on an ambitious project:

EvolvingAgents: Bio-Inspired Design and Evolution for LLM-Based Multi-Agent Systems

We’re building a two-pillar bio-inspired framework that couples within-task coordination with across-task organizational adaptation, and a long-horizon arena where civilizations evolve under limited resources, seasonal changes, and stochastic disasters.

Core Ideas:

• Stigmergic memory, typed reflex signaling, and an LLM-driven adaptive immune pathway for coordination and recovery
• A Meta-Evolver that turns recurrent experience into auditable runtime components with lineage, ratings, promotion, and retirement
• Group-level performance as the main selection signal

My Role:

• Proposed the civilization-style benchmark vision and led the Nature Environment scenario and evaluation design
• Coordinate interfaces across simulation, observation, memory, evolution, and analysis modules

A fascinating intersection of biology, AI, and multi-agent systems! 🌍🤖