paper-with-me

홈 › Papers

Sustained Gradient Alignment Mediates Subliminal Learning in a Multi-Step Setting: Evidence from MNIST Auxiliary Logit Distillation Experiment

2026-04-28 · Chayanon Kitkana, Shivam Arora arxiv

In the MNIST auxiliary logit distillation experiment, a student can acquire an unintended teacher trait despite distilling only on no-class logits through a phenomenon called subliminal learning. Under a single-step gradient descent assumption, subliminal learning theory attributes this effect to alignment between the trait and distillation gradients, but does not guarantee that this alignment persists in a multi-step setting. We empirically show that gradient alignment remains weakly but consistently positive throughout training and causally contributes to trait acquisition. We show that a mitigation method called liminal training works by attenuating the alignment and fails to stop trait acquisition in this setup. These results suggest that mitigation methods that operate in this regime may not reliably suppress trait acquisition when the first-order drive dominates.

📄 PDF Abstract BibTeX arXiv:2604.25779

Code (0)

등록된 구현이 없습니다.

Similar Papers 제목 키워드 기반

Thought Virus: Viral Misalignment via Subliminal Prompting in Multi-Agent Systems

2026-02-23 · Moritz Weckbecker, Jonas Müller, Ben Hagag, Michael Mulet arxiv

Subliminal prompting is a phenomenon in which language models are biased towards certain concepts or traits through prompting with semantically unrelated tokens. While prior work has examined subliminal prompting in user…

Subliminal Learning Is Steering Vector Distillation

2026-05-31 · Camila Blank, Agam Bhatia, Senthooran Rajamanoharan, Arthur Conmy 외 arxiv

Subliminal learning refers to a student language model acquiring a teacher's traits (e.g. a system-prompted preference for owls) when fine-tuned on the teacher's outputs, despite the outputs being semantically unrelated …

Subliminal Corruption: Mechanisms, Thresholds, and Interpretability

2025-10-22 · Reya Vir, Sarvesh Bhatnagar arxiv

As machine learning models are increasingly fine-tuned on synthetic data, there is a critical risk of subtle misalignments spreading through interconnected AI systems. This paper investigates subliminal corruption, which…

Emergent and Subliminal Misalignment Through the Lens of Data-Mediated Transfer

2026-05-12 · Baris Askin, Muhammed Ustaomeroglu, Anupam Nayak, Gauri Joshi 외 arxiv

Fine-tuning LLMs on narrow harmful datasets can induce Emergent Misalignment (EM), where models exhibit misaligned behavior far beyond the fine-tuning distribution. We argue that emergent misalignment can be better under…

Subliminal Learning is Non-Semantic Distillation

2026-08-06 · Ethan Hadley, Eren Gultepe arxiv

Subliminal Learning (SL) is a surprising type of generalization displayed by modern language models. It allows the transfer of a bias or behavior from a teacher model to a student by distilling from seemingly unrelated o…