Agent Capability
Agent capability research focuses on evaluating and enhancing the performance of artificial intelligence agents across diverse tasks, aiming to create more robust, adaptable, and reliable systems. Current research emphasizes developing novel agent architectures, such as those incorporating self-improvement mechanisms or normative modules, and improving evaluation methods through techniques like value function decomposition and benchmark creation tailored to specific domains (e.g., cybersecurity, biomedical science). These advancements are crucial for mitigating risks associated with increasingly capable AI systems and for advancing the broader field of artificial intelligence through more rigorous evaluation and improved agent design.
Papers
Adaptive In-conversation Team Building for Language Model Agents
Linxin Song, Jiale Liu, Jieyu Zhang, Shaokun Zhang, Ao Luo, Shijian Wang, Qingyun Wu, Chi Wang
Normative Modules: A Generative Agent Architecture for Learning Norms that Supports Multi-Agent Cooperation
Atrisha Sarkar, Andrei Ioan Muresanu, Carter Blair, Aaryam Sharma, Rakshit S Trivedi, Gillian K Hadfield