Tootfinder

Innocence in the Crossfire: Roles of Skip Connections in Jailbreaking Visual Language Models
Palash Nandi, Maithili Joshi, Tanmoy Chakraborty
https://arxiv.org/abs/2507.13761

Innocence in the Crossfire: Roles of Skip Connections in Jailbreaking Visual Language Models
Language models are highly sensitive to prompt formulations - small changes in input can drastically alter their output. This raises a critical question: To what extent can prompt sensitivity be exploited to generate inapt content? In this paper, we investigate how discrete components of prompt design influence the generation of inappropriate content in Visual Language Models (VLMs). Specifically, we analyze the impact of three key factors on successful jailbreaks: (a) the inclusion of detailed…

Ein paar Eindrücke einer Palästina Demonstration in Hannover bei der Ich mit der Linksjugend ['solid] teilgenommen habe.

Retrieval-Augmented Generation in Industry: An Interview Study on Use Cases, Requirements, Challenges, and Evaluation
Lorenz Brehme, Benedikt Dornauer, Thomas Str\"ohle, Maximilian Ehrhart, Ruth Breu
https://arxiv.org/abs/2508.14066

Retrieval-Augmented Generation in Industry: An Interview Study on Use Cases, Requirements, Challenges, and Evaluation
Retrieval-Augmented Generation (RAG) is a well-established and rapidly evolving field within AI that enhances the outputs of large language models by integrating relevant information retrieved from external knowledge sources. While industry adoption of RAG is now beginning, there is a significant lack of research on its practical application in industrial contexts. To address this gap, we conducted a semistructured interview study with 13 industry practitioners to explore the current state of R…

Learning in Context: Personalizing Educational Content with Large Language Models to Enhance Student Learning
Joy Jia Yin Lim, Daniel Zhang-Li, Jifan Yu, Xin Cong, Ye He, Zhiyuan Liu, Huiqin Liu, Lei Hou, Juanzi Li, Bin Xu
https://arxiv.org/abs/2509.15068

Learning in Context: Personalizing Educational Content with Large Language Models to Enhance Student Learning
Standardized, one-size-fits-all educational content often fails to connect with students' individual backgrounds and interests, leading to disengagement and a perceived lack of relevance. To address this challenge, we introduce PAGE, a novel framework that leverages large language models (LLMs) to automatically personalize educational materials by adapting them to each student's unique context, such as their major and personal interests. To validate our approach, we deployed PAGE in a semester-…

Scrub It Out! Erasing Sensitive Memorization in Code Language Models via Machine Unlearning
Zhaoyang Chu, Yao Wan, Zhikun Zhang, Di Wang, Zhou Yang, Hongyu Zhang, Pan Zhou, Xuanhua Shi, Hai Jin, David Lo
https://arxiv.org/abs/2509.13755

Scrub It Out! Erasing Sensitive Memorization in Code Language Models via Machine Unlearning
While Code Language Models (CLMs) have demonstrated superior performance in software engineering tasks such as code generation and summarization, recent empirical studies reveal a critical privacy vulnerability: these models exhibit unintended memorization of sensitive training data, enabling verbatim reproduction of confidential information when specifically prompted. To address this issue, several approaches, including training data de-duplication and differential privacy augmentation, have b…

FedMentor: Domain-Aware Differential Privacy for Heterogeneous Federated LLMs in Mental Health
Nobin Sarwar, Shubhashis Roy Dipta
https://arxiv.org/abs/2509.14275 https://

FedMentor: Domain-Aware Differential Privacy for Heterogeneous Federated LLMs in Mental Health
Privacy-preserving adaptation of Large Language Models (LLMs) in sensitive domains (e.g., mental health) requires balancing strict confidentiality with model utility and safety. We propose FedMentor, a federated fine-tuning framework that integrates Low-Rank Adaptation (LoRA) and domain-aware Differential Privacy (DP) to meet per-domain privacy budgets while maintaining performance. Each client (domain) applies a custom DP noise scale proportional to its data sensitivity, and the server adaptiv…

Large Language Model probabilities cannot distinguish between possible and impossible language
Evelina Leivada, Raquel Montero, Paolo Morosi, Natalia Moskvina, Tamara Serrano, Marcel Aguilar, Fritz Guenther
https://arxiv.org/abs/2509.15114

Large Language Model probabilities cannot distinguish between possible and impossible language
A controversial test for Large Language Models concerns the ability to discern possible from impossible language. While some evidence attests to the models' sensitivity to what crosses the limits of grammatically impossible language, this evidence has been contested on the grounds of the soundness of the testing material. We use model-internal representations to tap directly into the way Large Language Models represent the 'grammatical-ungrammatical' distinction. In a novel benchmark, we elicit…

Beyond Human Judgment: A Bayesian Evaluation of LLMs' Moral Values Understanding
Maciej Skorski, Alina Landowska
https://arxiv.org/abs/2508.13804 https://

Beyond Human Judgment: A Bayesian Evaluation of LLMs' Moral Values Understanding
How do large language models understand moral dimensions compared to humans? This first large-scale Bayesian evaluation of market-leading language models provides the answer. In contrast to prior work using deterministic ground truth (majority or inclusion rules), we model annotator disagreements to capture both aleatoric uncertainty (inherent human disagreement) and epistemic uncertainty (model domain sensitivity). We evaluate top language models (Claude Sonnet 4, DeepSeek-V3, Llama 4 Maveri…

The Sum Leaks More Than Its Parts: Compositional Privacy Risks and Mitigations in Multi-Agent Collaboration
Vaidehi Patil, Elias Stengel-Eskin, Mohit Bansal
https://arxiv.org/abs/2509.14284

The Sum Leaks More Than Its Parts: Compositional Privacy Risks and Mitigations in Multi-Agent Collaboration
As large language models (LLMs) become integral to multi-agent systems, new privacy risks emerge that extend beyond memorization, direct inference, or single-turn evaluations. In particular, seemingly innocuous responses, when composed across interactions, can cumulatively enable adversaries to recover sensitive information, a phenomenon we term compositional privacy leakage. We present the first systematic study of such compositional privacy leaks and possible mitigation methods in multi-agent…

Word Meanings in Transformer Language Models
Jumbly Grindrod, Peter Grindrod
https://arxiv.org/abs/2508.12863 https://arxiv.org/pdf/2508.12863

Word Meanings in Transformer Language Models
We investigate how word meanings are represented in the transformer language models. Specifically, we focus on whether transformer models employ something analogous to a lexical store - where each word has an entry that contains semantic information. To do this, we extracted the token embedding space of RoBERTa-base and k-means clustered it into 200 clusters. In our first study, we then manually inspected the resultant clusters to consider whether they are sensitive to semantic information. In …