The paper introduces GAPS, a dimension‑level gating approach for activation steering in language models. GAPS uses two training‑free gates—a static separability gate based on AUROC and a dynamic posterior gate based on a Gaussian model—to selectively apply steering vectors only to neurons that carry reliable concept information or are currently mis‑activated. Experiments on Gemma‑3 and Qwen‑3 show that GAPS improves or matches the performance of token‑level methods, notably reducing Gemma‑3’s toxicity rate from 6.52% to 0.48% under a fixed capability budget.
By Moghis Fereidouni, Muhammad Umair Haider, Hassan Sajjad, A. B. Siddique
arXiv:2605. 03058v2 Announce Type: replace-cross Abstract: A central goal of explainable AI is to express large language model (LLM) decision logic symbolically and ground it in internal mechanisms.
By Francesco Sovrano, Gabriele Dominici, Marc Langheinrich
arXiv:2607. 18639v1 Announce Type: new Abstract: Safety interventions on dual-use knowledge typically choose between destroying hazardous content (e.
By Seunghyun Lee, Dongyoon Han, Sangdoo Yun
arXiv:2607. 25907v1 Announce Type: cross Abstract: Activation steering controls model behavior by editing internal activations at inference time.
By Deepanshu Mody, Samarth Agarwal, Utkarsh Mittal, Dipesh Mahato
arXiv:2608. 14392v1 Announce Type: new Abstract: Neuron- and path-level interventions offer the finest-grained route to defending large language models (LLMs) against jailbreak attacks, yet existing methods fall short of this promise, i.
By Wei Zhao, Zhe Li, Peixin Zhang, Jun Sun
Safety interventions on dual-use knowledge typically choose between destroying hazardous content (e. g.
arXiv:2606. 08682v1 Announce Type: cross Abstract: Activation steering has emerged as a popular inference-time technique for modulating the behavior of large language models (LLMs).
By Qi Cao, Jian Lou, Meiting Liu, Wenjie Feng, Dan Li, See-Kiong Ng, Anh Tuan Luu
arXiv:2602. 06941v2 Announce Type: replace-cross Abstract: Large language models can recover mid-generation from task-misaligned activation steering, producing explicit verbal restarts (e.
By Alex McKenzie, Keenan Pepper, Stijn Servaes, Martin Leitgab, Murat Cubuktepe, Mike Vaiana, Diogo de Lucena, Judd Rosenblatt, Michael S. A. Graziano
arXiv:2608. 15687v1 Announce Type: new Abstract: Sycophancy, the tendency of a language model to change its answer to match a user's stated belief, is a common alignment failure.
By Kareem Hassani, Chaymaa Abbas, Lama Mawlawi, Mariette Awad
The paper introduces a probe‑free method called the Neuron Separability Index (NSI) to assess how individual neurons in large language models distinguish grammatical from ungrammatical sentences using linguistic minimal pairs. Across 68 linguistic paradigms and seven model checkpoints, the study finds that while raw separability for morphology and syntax peaks early, single‑unit selectivity is sparse and weak, with rare strongly selective "grandmother neurons." Moreover, the research shows a dissociation between whole‑vector linear separability, single‑neuron selectivity, and behavioral competence, and demonstrates that targeted ablations can further separate activation selectivity from causal reliance.
arXiv:2608. 10214v1 Announce Type: new Abstract: Do large language models contain domain-specific parametric shells: concentrated, causally necessary neuron populations whose removal selectively degrades a target domain while sparing others?
By Marcus Armstrong, Navid Ayoobi, Arjun Mukherjee
The paper introduces Exemplar Partitioning (EP), an unsupervised technique that constructs interpretable feature dictionaries from large language model activations by clustering streamed activations into Voronoi regions defined by exemplars and their averages. EP allows comparison of dictionaries across layers, checkpoints, and architectures, and demonstrates utility in interpreting model behavior, tracking training dynamics, detecting hidden concepts, and enabling targeted interventions. Experiments on Gemma‑2‑2B and Llama‑3.1‑8B show EP can reveal how instruction tuning reorganizes harmful prompt activations, facilitate interventions that alter model responses, and achieve high concept‑detection performance while requiring far fewer construction tokens than comparable methods.
By Jessica Rumbelow