The paper investigates whether fine‑tuning a language model erases previously embedded activation steering interventions that suppress refusals and encourage brevity. Across five instruction‑tuned models (3B–14B) subjected to non‑adversarial supervised fine‑tuning (SFT) and reinforcement learning from human feedback (RLHF), the authors find that the steering’s behavioural effect degrades when the fine‑tuning objective conflicts with the targeted behaviour, yet the underlying weight edits remain largely unchanged. Mechanistically, the steering vectors survive with minimal alteration, but functionally the steering is vulnerable and must be re‑validated after downstream training.
By Philipp E. Glass, Allan Tucker, Yongmin Li, Alina Miron
arXiv:2607. 18114v1 Announce Type: cross Abstract: Modern LLMs are alarmingly susceptible to surprisingly simple immaterial changes of input prompts: a casual hint, an incorrectly labeled few-shot example, or a fake prior assistant turn often flips an originally correct answer.
By Prakhar Gupta, Terry Jingchen Zhang, Florent Draye, Bernhard Sch\"olkopf, Zhijing Jin
arXiv:2608. 04347v1 Announce Type: new Abstract: Fine-tuning enables a source model to acquire desired capabilities and behaviors in a target domain while retaining much of its general-purpose competence.
By Kotaro Yoshida, Laura Gomezjurado Gonzalez, Yukinori Yamamoto, Yuji Naraki, Ryotaro Shimizu, Wenya Wang
Modern LLMs are alarmingly susceptible to surprisingly simple immaterial changes of input prompts: a casual hint, an incorrectly labeled few-shot example, or a fake prior assistant turn often flips an originally correct answer. We study where this susceptibility, spanning sycophancy and related cue-induced biases, lives inside the model.
arXiv:2605.01699v4 Announce Type: replace
Abstract: Recent attacks show that behavioural unlearning of large language models leaves internal traces recoverable by adversarial probes. We characterise...
By Anamika Paul Rupa, Anietie Andy
arXiv:2608. 15483v1 Announce Type: new Abstract: Modern deep networks are trained through long update trajectories, yet their temporal organization remains less systematically characterized than architectures, losses, or optimizers.
By Fanqi Wang, Weisheng Tang, Hairong Qi
arXiv:2606. 31591v1 Announce Type: cross Abstract: Emergent misalignment (EM) is a recently discovered phenomenon in LLMs where fine-tuning on a narrow misaligned task, such as writing insecure code, leads to broadly misaligned behaviour on unrelated prompts.
By Jason R. Brown, Patrick Leask, Lev McKinney
arXiv:2606. 07631v1 Announce Type: cross Abstract: Emergent misalignment (EM) occurs when narrow finetuning causes a model to behave dangerously outside the finetuning task.
By Huy Nghiem, Sy-Tuyen Ho, Sarah Wiegreffe, Hal Daum\'e III
arXiv:2606. 01060v1 Announce Type: cross Abstract: Preference alignment has substantially improved the observable behavior of large language models, yet it remains unclear what alignment changes internally.
By Partha Pratim Saha, Samarth Raina, Mayur Parvatikar, Amit Dhanda, Vinija Jain, Aman Chadha, Amitava Das
Modern deep networks are trained through long update trajectories, yet their temporal organization remains less systematically characterized than architectures, losses, or optimizers. We study short-h...
The paper investigates what a steerable valence direction in large language models (LLMs) actually represents, focusing on a good‑bad outcome direction in a maze task. By using controlled interventions that separate the realized outcome from the informational history that led to it, the authors find that directions trained on one explicit outcome encoding transfer well to another, suggesting the readout is not tied to surface form. However, when the same outcome is achieved through announced versus unannounced histories, transfer performance drops sharply, indicating that the post‑event readout remains strongly conditioned on the earlier announcement. In a matched maze‑reinforcement‑learning run, the post‑RL direction becomes more predictive of reference‑MDP return and the policy depends more on it, yet the history dependence persists. These findings support a functional, value‑related interpretation of the direction but argue against identifying it with a history‑invariant scalar valence state.
By Weihan Li, Xinlei Chen, Yuhan Song, Xiaofeng Lin, Tianshi Zheng
arXiv:2607. 25063v1 Announce Type: new Abstract: Developers judge a model checkpoint by how it behaves.
By Cen Lu, Yung-Chen Tang, Andrea Cavallaro