Daily Research Brief 2026-05-24

📊 Token usage: estimated from retrieval and writing scale.

Covers the latest AI research, open source and industry moves, updated daily.

Editor’s Note

The paper thread today: a survey of multi-agent systems beyond individual intelligence, trust-region fine-tuning for agent collaboration (TeamTR), participatory AI mapping, NeuroMAS (multi-agent systems as neural networks), and a unified Theory of Agent.

1. Latest arXiv Papers

1. Beyond Individual Intelligence: A Survey of Multi-Agent Systems and Collaboration — https://arxiv.org/abs/2605.14892

2. TeamTR: Trust-Region Fine-Tuning for Multi-Agent LLM Collaboration — https://arxiv.org/abs/2605.15207

3. Voices in the Loop: Mapping Participatory AI — https://arxiv.org/abs/2605.16827

4. Multi-Paradigm Agent Interaction in Practice: A Systematic Study — https://arxiv.org/abs/2605.16821

5. NeuroMAS: Multi-Agent Systems as Neural Networks with Joint Optimization — https://arxiv.org/abs/2605.16757

6. Autonomous Discovery of Neural Architectures by AI Agents: AIRA-Compose and AIRA-Design

7. Theory of Agent (ToA): A Unified Framework for Agent Intelligence — https://arxiv.org/abs/2506.00886

8. MoE Pre-Training: Dynamic Decomposition of Neurons

9. Nudging Beyond the Comfort Zone: Efficient Strategies for Agent Exploration — https://arxiv.org/abs/2605.15726

10. CODA: Rewriting Transformer Blocks as GEMM-Epilogue Units