arXiv:2602. 09533v2 Announce Type: replace Abstract: Direct preference optimization (DPO) has emerged as a promising approach for aligning large language models (LLMs) with human preferences.
By Masanari Oi, Mahiro Ukai, Masahiro Kaneko, Naoaki Okazaki, Nakamasa Inoue
The paper argues that language models’ intransitive preferences arise from multiple internally consistent latent orderings rather than noise around a single ordering. By demonstrating that a single ordering cannot explain observed inconsistencies and introducing a noise‑augmented mixture Bradley‑Terry model, the authors show that mixtures of orderings better capture preference structure across several models and tasks. A case study on Moral Machine dilemmas further illustrates that models can share latent components even when aggregate preferences differ.
By Aviral Chawla, William H. W. Thompson, Jean-Gabriel Young
arXiv:2410. 15595v4 Announce Type: replace Abstract: With the rapid advancement of large language models (LLMs), aligning policy models with human preferences has become increasingly critical.
By Wenyi Xiao, Zechuan Wang, Leilei Gan, Shuai Zhao, Zongrui Li, Ruirui Lei, Wanggui He, Luu Anh Tuan, Long Chen, Hao Jiang, Zhou Zhao, Fei Wu
arXiv:2606. 19744v1 Announce Type: cross Abstract: Aligning language models with human preferences often requires optimising multiple behavioural objectives.
By Pranav Bhandari, Nicolas Fay, Amitava Datta, Usman Naseem, Mehwish Nasim
The paper introduces DSPA, a dynamic sparse autoencoder (SAE) steering technique that aligns language model outputs with user preferences during inference, avoiding costly weight updates. DSPA constructs a conditional-difference map from preference triples to adjust token-active latents, improving MT‑Bench scores and matching AlpacaEval performance on models like Gemma‑2 and Qwen3 while preserving accuracy. It demonstrates robustness with limited preference data, outperforms the two‑stage RAHF‑SCIT pipeline in FLOPs, and reveals that preference directions are largely driven by discourse and stylistic cues.
By James Wedgwood, Aashiq Muhamed, Mona T. Diab, Virginia Smith
The paper introduces BALIGN, a balanced data selection strategy designed to reduce catastrophic forgetting—referred to as the alignment tax—in large language models during preference-based alignment. By analyzing preference optimization gradients, the authors identify three data-centric features that influence parameter drift: the reference model's log-probability margin, token length differences between chosen and rejected responses, and TF‑IDF similarity to general capability corpora. BALIGN aggregates these features into a composite risk score to filter out high-risk preference samples, thereby preserving foundational capabilities while maintaining alignment gains with minimal computational overhead.
By Minsu Kim, Jianxun Lian, Xing Xie, Steven Euijong Whang