Constructing Challenging Browser-Use Tasks by Controlled Environment Interventions
This paper evaluates six strong browser-use agents, three GUI-only agents that see only screenshots, and humans, evaluating the construction of challenging instances from tasks agents already solve, turning difficulty into a programmable property of the environment.
Xun-Jian Yin, Tian-Chen Guan, Jin-Ao Wang et al.
· 0 citations