Continual Learning Circle Meetup & Dinner

​Modern AI systems are remarkably capable, but they still struggle to consistently generate genuinely new ideas, strategies, and behaviors. Instead, they often fall into local optima, repeating familiar patterns rather than exploring alternative approaches. This remains a major obstacle for continual learning and self-improvement.

​Jonathan Li (visiting us from Caltech) will discuss how search can drive exploration and open-ended discovery in large language models and AI agents. He is also a founding member of Asari AI, and a researcher at NEC Laboratories America, with work published at ICLR, NeurIPS, and ICML.

​This session explores recent work on structured, language-native search methods such as DISC and SFS, which enable agents to explore diverse reasoning paths, strategies, and solutions beyond what is directly represented in their training data. It will also examine how these methods can support agents that learn continuously from experience, including a case study on increasingly sophisticated social reasoning, coordination, and strategy in multi-agent games. Together, these approaches point toward AI systems that can autonomously acquire new skills, adapt to changing environments, and self-improve through experience.

​This event is part of the Frontier AI Research Series hosted by Across AI, AI Circle, and Adaption on the future of grounding, memory, reasoning, and continual learning in AI systems.