arXiv:2607. 16240v1 Announce Type: cross Abstract: Direct Alignment Algorithms (DAAs) such as DPO have become a common way to post-train and align LLMs with human preferences.
By Shawn Im, Federico Danieli, Skyler Seto, Barry-John Theobald, Katherine Metcalf
The study investigates whether coefficient priors derived from large language models (LLMs) can enhance out‑of‑distribution (OOD) tabular prediction by serving as a shrinkage target in regularised regression. Across five datasets, LLM‑elicited priors improved OOD mean squared error on four datasets by up to 7.9% and yielded better probabilistic accuracy in classification tasks, though the gains varied and were sometimes comparable to in‑domain improvements. Cross‑validation was used to select priors, but the variability among candidates meant that the method mainly avoided the worst priors rather than consistently finding the best.
"whyItMatters":"The findings suggest that LLM‑derived priors can provide a scalable alternative to expert elicitation for improving model robustness to distribution shifts, though their effectiveness depends on dataset size and prior variability."
By S{\o}lve Langseth, Kristoffer Herland Hellton, Hugo L. Hammer
arXiv:2609.22215v1 Announce Type: cross
Abstract: Knowledge distillation can transmit unintended behavioral traits from a teacher model to a student through training data that appear semantically unr...
By Atsushi Yanagisawa, Brendan Gho, Rajendran Ramesh Babu Manoj Narender, Kevin Zhu, Madhur Panwar, Antonio Mari
Reference-Guided Machine Unlearning (ReGUn) is a vision unlearning framework that prioritizes distributional indistinguishability over degradation-based heuristics. It uses disjoint held-out data to create a class-conditioned reference distribution for distillation, guiding forget samples toward non-member behavior without explicitly degrading predictions. Experiments across various architectures and datasets show that ReGUn achieves a competitive forgetting–utility trade-off and closely matches retrain-like membership inference behavior.
By Jonas Mirlach, Sonia Laguna, Julia E. Vogt
The paper investigates how scaling the amount of model-generated, off‑task distillation data affects the recoverability of teacher‑induced traits in student models. In a controlled subliminal‑learning setup, teachers are prompted to express a target trait, producing restricted data such as number‑only completions. Students trained on larger independent datasets show a clearer manifestation of the teacher’s trait in a separate evaluation domain, with the effect being strongest when the trait is already favored or when alternative traits are present. The authors find this trend holds across model families, trait types, multi‑trait settings, and cross‑model transfer, and suggest that scaling should be coupled with trait‑aware curation and evaluation.
By Zhichen Dong, Zhixuan Liu, Yuyu Fan, Xiangtian Li, Shuyang Zhang, Chao Yang
The paper investigates subliminal learning, where large language models inherit unintended traits from data generated by other models. It compares hard‑label distillation (students learn from biased teacher outputs) and on‑policy soft‑label distillation (students generate data while teacher logits guide learning), finding that transfer persists across various training conditions and that teacher‑student roles can be asymmetrically effective. Synthetic experiments show that preference signals can be transmitted even without prompt dependence, and that removing certain tokens reduces but does not eliminate transfer, with soft‑label distillation yielding more consistent results with less training.
By Angela Tang, Vishesh Anand, Gon\c{c}alo Paulo