Beyond the Surface: Probing the Ideological Depth of Large Language Models

Safety-aligned large language models (LLMs) are increasingly used in sensitive applications where fairness is crucial. Evaluating their fairness is complex, often relying on standard question-answer methods that misinterpret refusal responses as indicators of fairness. This paper introduces the concept of silenced biases, which are unfair preferences hidden within the models' latent space, masked by safety-alignment. Previous methods have limitations, prompting the need for new approaches to uncover these biases effectively.

The paper titled 'Fair In-Context Learning via Latent Concept Variables' explores the in-context learning (ICL) capabilities of large language models (LLMs) in handling tabular data. It highlights the potential for LLMs to inherit biases from pre-training data, which can lead to discrimination in high-stakes applications. The authors propose an optimal demonstration selection method using latent concept variables to enhance task adaptation and fairness, alongside data augmentation strategies to minimize correlations between sensitive variables and predictive outcomes.

The paper titled 'Modeling and Predicting Multi-Turn Answer Instability in Large Language Models' discusses the evaluation of large language models (LLMs) in terms of their robustness during user interactions. The study employs multi-turn follow-up prompts to assess changes in model answers and accuracy dynamics using Markov chains. Results indicate vulnerabilities in LLMs, with a 10% accuracy drop for Gemini 1.5 Flash after a 'Think again' prompt over nine turns, and a 7.5% drop for Claude 3.5 Haiku with a reworded question. The findings suggest that accuracy can be modeled over time.

A recent study published on arXiv examines the phenomenon of negative bias in large language models (LLMs), which refers to their tendency to generate negative responses in binary decision tasks. The research highlights that previous studies have primarily focused on identifying negative attention heads that contribute to this bias. The authors introduce a new evaluation pipeline that categorizes responses based on the model's parametric knowledge, revealing that the format of prompts significantly influences the responses more than the semantics of the content itself.

The paper titled 'Identifying and Analyzing Performance-Critical Tokens in Large Language Models' explores how large language models (LLMs) utilize in-context learning (ICL) for few-shot learning. It categorizes tokens in ICL prompts into content, stopword, and template tokens, aiming to identify those that significantly impact LLM performance. The study reveals that template and stopword tokens have a greater influence on performance than informative content tokens, challenging existing assumptions about human attention to informative words.

Recent advancements in large language models (LLMs) highlight their potential in automating scientific research ideation. Current methods often produce ideas that do not meet expert standards of novelty, feasibility, and effectiveness. To address these issues, a new framework is proposed that combines Supervised Fine-Tuning (SFT) and controllable Reinforcement Learning (RL) to enhance the quality of generated research ideas through a two-stage approach.

Large language models (LLMs) have shown remarkable capabilities in generation tasks but often produce harmful or biased content, raising ethical concerns. To address these issues, representation engineering has been proposed as a method to guide model behavior through the use of steering vectors. However, existing methods face challenges such as limited control and content quality degradation. The proposed Sparse Representation Steering (SRS) framework aims to provide fine-grained and interpretable control over LLM behavior by disentangling internal activations.

The paper titled 'Pre-Attention Expert Prediction and Prefetching for Mixture-of-Experts Large Language Models' introduces a method to enhance the efficiency of Mixture-of-Experts (MoE) Large Language Models (LLMs). The authors propose a pre-attention expert prediction technique that improves accuracy and reduces computational overhead by utilizing activations before the attention block. This approach aims to optimize expert prefetching, achieving about a 15% improvement in accuracy over existing methods.