UniAPL: A Unified Adversarial Preference Learning Framework for Instruct-Following
This paper introduces the UniAPL framework to enhance alignment in large language models (LLMs) via unified preference learning, addressing distributional mismatch issues in traditional approaches.
Paired by the Teacher: Turning Unpaired Data into High-Fidelity Pairs for Low-Resource Text Generation
This paper presents PbT, a novel method using teacher-student modeling to create high-quality paired training data for low-resource NLG tasks, significantly improving performance with limited data.
Investigating Language and Retrieval Bias in Multilingual Previously Fact-Checked Claim Detection
The study reviews language and retrieval biases in multilingual LLMs for fact-checking tasks, revealing disparities in performance based on language and model characteristics.
Knowledge Extraction on Semi-Structured Content: Does It Remain Relevant for Question Answering in the Era of LLMs?
This work investigates the relevance of knowledge extraction methods in enhancing web-based QA systems’ performance amidst the emergence of LLMs.
Scaling with Collapse: Efficient and Predictable Training of LLM Families
The paper explores LLM training consistency and presents a method that improves efficiency and predictability in scaling model families, potentially reducing training costs.
Towards Trustworthy Lexical Simplification: Exploring Safety and Efficiency with Small LLMs
The research introduces a framework leveraging small LLMs for lexical simplification, emphasizing output safety and efficiency for vulnerable user groups.
Surveying Perceptions of AI-Enhanced Document Assistance: A Multinational Perspective
This survey presents insights from a global audience on the perceived benefits and risks associated with AI-driven document assistance tools across various regions.
Hyperdimensional Probe: Decoding LLM Representations via Vector Symbolic Architectures
The study introduces a new paradigm for decoding information from LLMs by employing Vector Symbolic Architectures to enhance interpretability.
From Human Annotation to Automation: LLM-in-the-Loop Active Learning for Arabic Sentiment Analysis
The paper proposes a framework for active learning in Arabic sentiment analysis, utilizing LLMs to assist in annotation and demonstrating efficient labeling strategies.
Speculative Verification: Exploiting Information Gain to Refine Speculative Decoding
This work presents Speculative Verification, enhancing speculative decoding by dynamically predicting speculation accuracy to optimize LLM inference efficiency.