Continual Test-Time Learning with Evolution Strategies

Guangya Hao, Yiting Duan, Longyan Tan +5

Large language models increasingly operate in open-ended environments, where they are expected to continually adapt to evolving tasks and distributions, often without access to ext…