2 papers
cs.CL2026
Causal Tongue-Tie: LLMs Can Encode Causal Direction, But Their Yes/No Outputs Fail to Express
Ziyi Ding, Xiao-Ping Zhang
We find a mismatch between what large language models encode about a causal question and what they answer. On anti-commonsense CLadder items, a fixed linear probe recovers the evid…
cs.LG2026
CausalVAE as a Plug-in for World Models: Towards Reliable Counterfactual Dynamics
Ziyi Ding, Xianxin Lai, Weiyu Chen +2
In this work, CausalVAE is introduced as a plug-in structural module for latent world models and is attached to diverse encoder-transition backbones. Across the reported benchmarks…