The authors extended C-WAH (household collaboration) and CalBench (calendar scheduling) by introducing asymmetric information: each agent has personal observations and communication histories, while the original goals, constraints, and validators are preserved.
Model toM accuracy on stories does not predict planning in tasks
A multi-level evaluation shows that language models can correctly infer beliefs but fail to use them in verifiable collaboration and scheduling tasks.
Chinese Tech
Zhenshuai Yin · Yongxin Guo · Mingda Zhang · Zhongze Luo · Haoyue Liu · Lianmin Zhang · +1 more
The Chinese University of Hong Kong (Shenzhen) · ByteDance Inc. · Shenzhen Research Institute of Big Data
Research Digest··1 min read
Yin et al. introduce a multi-level evaluation of theory of mind that separates belief inference from its use in decision-making and task execution.
The Daily Edition
The front page in your inbox each weekday morning: the lead, the picks, and why each one earned its place.