A Multilingual, Large-Scale Study of the Interplay between LLM Safeguards, Personalisation, and Disinformation

arXiv — cs.CL•Thursday, October 30, 2025 at 4:00:00 AM

A recent study explores how large language models (LLMs) can generate personalized disinformation across different languages and demographics. This large-scale, multilingual analysis is significant as it sheds light on the capabilities of LLMs in creating targeted false narratives, which is crucial for understanding the potential risks and implications of AI in information dissemination. By employing a red teaming methodology, the research prompts eight advanced LLMs with various false narratives and demographic personas, highlighting the need for robust safeguards against misuse.

— via World Pulse Now AI Editorial System

Read Original

Was this article worth reading? Share it

Fyli

AI-powered customer discovery tool to instantly identify your target audience.

AI & DataTry the app

mailgo

Generate accurate email leads from names and domains using advanced pattern recognition.

Marketing & CommerceTry the app

DataReclaimer

Extract LinkedIn data to enhance and scale your outreach campaigns effectively.

Marketing & CommerceTry the app

Continue Readings

arXiv — cs.CLa day ago

RhinoInsight: Improving Deep Research through Control Mechanisms for Model Behavior and Context

PositiveArtificial Intelligence

RhinoInsight has been introduced as a new framework aimed at enhancing deep research capabilities by incorporating control mechanisms that improve model behavior and context management. This framework addresses issues such as error accumulation and context rot, which are prevalent in existing linear pipelines used by large language models (LLMs). The two main components are a Verifiable Checklist module and an Evidence Audit module, which work together to ensure robustness and traceability in research outputs.

Read full article

via arXiv — cs.CL

arXiv — cs.CLa day ago

A Benchmark for Zero-Shot Belief Inference in Large Language Models

PositiveArtificial Intelligence

A new benchmark for zero-shot belief inference in large language models (LLMs) has been introduced, assessing their ability to predict individual stances on various topics using data from an online debate platform. This systematic evaluation highlights the influence of demographic context and prior beliefs on predictive accuracy.

Read full article

via arXiv — cs.CL

arXiv — cs.CLa day ago

Community-Aligned Behavior Under Uncertainty: Evidence of Epistemic Stance Transfer in LLMs

PositiveArtificial Intelligence

A recent study investigates how large language models (LLMs) aligned with specific online communities respond to uncertainty, revealing that these models exhibit consistent behavioral patterns reflective of their communities even when factual information is removed. This was tested using Russian-Ukrainian military discourse and U.S. partisan Twitter data.

Read full article

via arXiv — cs.CL

arXiv — cs.CLa day ago

Does Reinforcement Learning Really Incentivize Reasoning Capacity in LLMs Beyond the Base Model?

NeutralArtificial Intelligence

Recent research has critically evaluated the effectiveness of Reinforcement Learning with Verifiable Rewards (RLVR) in enhancing the reasoning capabilities of large language models (LLMs). The study found that while RLVR-trained models perform better than their base counterparts on certain tasks, they do not exhibit fundamentally new reasoning patterns, particularly at larger evaluation metrics like pass@k.

Read full article

via arXiv — cs.CL

arXiv — cs.CLa day ago

Principled Context Engineering for RAG: Statistical Guarantees via Conformal Prediction

PositiveArtificial Intelligence

A new study introduces a context engineering approach for Retrieval-Augmented Generation (RAG) that utilizes conformal prediction to enhance the accuracy of large language models (LLMs) by filtering out irrelevant content while maintaining relevant evidence. This method was tested on the NeuCLIR and RAGTIME datasets, demonstrating a significant reduction in retained context without compromising factual accuracy.

Read full article

via arXiv — cs.CL

arXiv — cs.CLa day ago

L2V-CoT: Cross-Modal Transfer of Chain-of-Thought Reasoning via Latent Intervention

PositiveArtificial Intelligence

Researchers have introduced L2V-CoT, a novel training-free approach that facilitates the transfer of Chain-of-Thought (CoT) reasoning from large language models (LLMs) to Vision-Language Models (VLMs) using Linear Artificial Tomography (LAT). This method addresses the challenges VLMs face in multi-step reasoning tasks due to limited multimodal reasoning data.

Read full article

via arXiv — cs.CL

arXiv — cs.CLa day ago

SGM: A Framework for Building Specification-Guided Moderation Filters

PositiveArtificial Intelligence

A new framework named Specification-Guided Moderation (SGM) has been introduced to enhance content moderation filters for large language models (LLMs). This framework allows for the automation of training data generation based on user-defined specifications, addressing the limitations of traditional safety-focused filters. SGM aims to provide scalable and application-specific alignment goals for LLMs.

Read full article

via arXiv — cs.CL

arXiv — cs.CLa day ago

HyperbolicRAG: Enhancing Retrieval-Augmented Generation with Hyperbolic Representations

PositiveArtificial Intelligence

HyperbolicRAG has been introduced as an innovative retrieval framework that enhances retrieval-augmented generation (RAG) by integrating hyperbolic geometry into graph-based approaches. This method aims to improve the representation of complex knowledge graphs by aligning semantic similarity with hierarchical depth, addressing limitations of traditional Euclidean embeddings.

Read full article

via arXiv — cs.CL