Mechanistic Interpretability Should Prioritize Feature Consistency in Sparse Autoencoders (2026 · 1 citation)
Sparse autoencoders (SAEs) are a popular way to break LLM activations into sparse, human-readable features.
Advancing Reasoning in Diffusion Language Models with Denoising Process Rewards (2026)
Masked diffusion language models (MdLLMs) such as LLaDA generate text by gradually unmasking a sequence. Reinforcement learning methods such as diffu-GRPO reward only the correctness of the final answer, so they do not supervise the denoising steps themselves. As a result, models often waste intermediate steps on…

All works

3 works of theirs in our library, the most cited first; 2 with a page here, the others link their original until we summarise them.

Mechanistic Interpretability Should Prioritize Feature Consistency in Sparse AutoencodersAnnual Meeting of the Association for Computational Linguistics (ACL)Xiangchen Song, Aashiq Muhamed, Yujia Zheng, et al.2026 · 1 citation
Advancing Reasoning in Diffusion Language Models with Denoising Process RewardsAnnual Meeting of the Association for Computational Linguistics (ACL)Shaoan Xie, Lingjing Kong, Xiangchen Song, et al.2026
From Reasoning Traces to Reusable Modules: Understanding Compositional Generalization in Language Model Reasoning ↗arXiv (Cornell University) · no summary yetLingjing Kong et al.2026

Co-authors here

8 co-authors with a page, the most papers together first.

Where their works here were published, the most works first, with the editions of those that have a page.

Papers of our library cite their works 1 time (of the 149,439 papers whose bibliographies we read).

Updated 2 November 2026