VLMs preserve linearly separable visual magnitudes and can compare them, yet collapse at symbolic mapping because visual and textual number spaces remain fractured and disjoint.
hub
Toxicity in chatgpt: Analyzing persona-assigned language models , url =
12 Pith papers cite this work, alongside 144 external citations. Polarity classification is still indexing.
12
Pith papers citing it
144
external citations · OpenAlex
hub tools
citation-role summary
background 3
citation-polarity summary
roles
background 3polarities
background 3representative citing papers
An AI-agent social platform generated mostly neutral content whose use in fine-tuning reduced model truthfulness comparably to human Reddit data, suggesting limited unique harm but flagging tail risks like secret leaks.
Composable LoRA adapters can amplify or suppress OCEAN traits in LLMs, combine approximately additively, preserve moderate-scale capability, and move safety-relevant behaviours.
Schützen is a German-Bulgarian LLM safety dataset showing pronounced cross-language differences in model safety behavior.
DISCA converts within-country disagreement among World Values Survey personas into a bounded logit correction that reduces cultural misalignment by 10-24% on MultiTP for models 3.8B and larger across 20 countries, without any weight updates.
Debiasing-DPO reduces bias to spurious social contexts by 84% and improves predictive accuracy by 52% on average for LLMs evaluating U.S. classroom transcripts.
LLMs show split alignment with human hate speech annotations (strong on explicit attributes, inverted on evaluative ones), and attribute-based ridge regression reconstructs continuous scores with R² up to 0.71.
LLMs assigned high or low status personas in multi-turn dialogues exhibit socio-cognitive effects including language coordination, pronoun patterns, persuasion success, and compliance with unsafe requests.
Monte Carlo simulations of LLM agents confirm that toxic debates take 25% longer to converge, with larger delays in smaller models, and show a first-mover advantage independent of toxicity.
The paper introduces a taxonomy of AI safety for LLMs organized into Trustworthy AI, Responsible AI, and Safe AI perspectives, accompanied by a review of state-of-the-art methods, challenges, and future directions.
Activation steering with FLORES-derived language vectors produces modest, layer-sensitive and language-dependent gains on cultural awareness tasks, with some settings degrading performance and strong interaction with prompt design.
A survey that catalogs threat models, detection approaches, and mitigation strategies for toxicity in multilingual LLMs while identifying challenges such as uneven language coverage and culturally variable harm definitions.
citing papers explorer
-
Dissociative Identity: Language Model Agents Lack Grounding for Reputation Mechanisms
VLMs preserve linearly separable visual magnitudes and can compare them, yet collapse at symbolic mapping because visual and textual number spaces remain fractured and disjoint.
-
The Moltbook Files: A Harmless Slopocalypse or Humanity's Last Experiment
An AI-agent social platform generated mostly neutral content whose use in fine-tuning reduced model truthfulness comparably to human Reddit data, suggesting limited unique harm but flagging tail risks like secret leaks.
-
Persona Cartography: Charting Language Model Personality Traits in Weight Space
Composable LoRA adapters can amplify or suppress OCEAN traits in LLMs, combine approximately additively, preserve moderate-scale capability, and move safety-relevant behaviours.
-
Sch\"utzen: Evaluating LLM Safety in Bulgarian and German Contexts
Schützen is a German-Bulgarian LLM safety dataset showing pronounced cross-language differences in model safety behavior.
-
Training-Free Cultural Alignment of Large Language Models via Persona Disagreement
DISCA converts within-country disagreement among World Values Survey personas into a bounded logit correction that reduces cultural misalignment by 10-24% on MultiTP for models 3.8B and larger across 20 countries, without any weight updates.
-
Mitigating LLM biases toward spurious social contexts using direct preference optimization
Debiasing-DPO reduces bias to spurious social contexts by 84% and improves predictive accuracy by 52% on average for LLMs evaluating U.S. classroom transcripts.
-
Attribute-Based Diagnosis of LLM Alignment with Hate Speech Annotations
LLMs show split alignment with human hate speech annotations (strong on explicit attributes, inverted on evaluative ones), and attribute-based ridge regression reconstructs continuous scores with R² up to 0.71.
-
Do LLM Agents Mirror Socio-Cognitive Effects in Power-Asymmetric Conversations?
LLMs assigned high or low status personas in multi-turn dialogues exhibit socio-cognitive effects including language coordination, pronoun patterns, persuasion success, and compliance with unsafe requests.
-
Beyond Inefficiency: Systemic Costs of Incivility in Multi-Agent Monte Carlo Simulations
Monte Carlo simulations of LLM agents confirm that toxic debates take 25% longer to converge, with larger delays in smaller models, and show a first-mover advantage independent of toxicity.
-
AI Safety Landscape for Large Language Models: Taxonomy, State-of-the-art, and Future Directions
The paper introduces a taxonomy of AI safety for LLMs organized into Trustworthy AI, Responsible AI, and Safe AI perspectives, accompanied by a review of state-of-the-art methods, challenges, and future directions.
-
DFKI-MLT at SemEval-2026 TASK 7: Steering Multilingual Models Towards Cultural Knowledge
Activation steering with FLORES-derived language vectors produces modest, layer-sensitive and language-dependent gains on cultural awareness tasks, with some settings degrading performance and strong interaction with prompt design.
-
A Survey of Toxicity Detection and Mitigation Strategies for Multilingual Language Models
A survey that catalogs threat models, detection approaches, and mitigation strategies for toxicity in multilingual LLMs while identifying challenges such as uneven language coverage and culturally variable harm definitions.