Harnessing the Unseen: The Hidden Influence of Intrinsic Knowledge in Long-Context Language Models

Recent advances in long-context models (LCMs), designed to handle extremely long input contexts, primarily focus on utilizing external contextual information, often leaving the influence of large language models' intrinsic knowledge underexplored. In this work, we investigate how this intrinsic knowledge affects content generation and demonstrate that its impact becomes increasingly pronounced as context length extends. Furthermore, we show that the model's ability to utilize intrinsic knowledge, which we call intrinsic retrieval ability, does not improve simultaneously with its ability to leverage contextual knowledge through extrinsic retrieval ability. Moreover, better extrinsic retrieval can interfere with the model's ability to use its own knowledge effectively, limiting its full potential. To bridge this gap, we design a simple yet effective Hybrid Needle-in-a-Haystack test that evaluates models based on their capabilities across both retrieval abilities, rather than solely emphasizing extrinsic retrieval ability. Our experimental results reveal that Qwen-2.5 models significantly outperform Llama-3.1 models, demonstrating superior intrinsic retrieval ability. Moreover, even the more powerful Llama-3.1-70B-Instruct model fails to exhibit better performance under LCM conditions, highlighting the importance of evaluating models from a dual-retrieval perspective.

ExpertGenQA: Open-ended QA generation in Specialized Domains

Generating high-quality question-answer pairs for specialized technical domains remains challenging, with existing approaches facing a tradeoff between leveraging expert examples and achieving topical diversity. We present ExpertGenQA, a protocol that combines few-shot learning with structured topic and style categorization to generate comprehensive domain-specific QA pairs. Using U.S. Federal Railroad Administration documents as a test bed, we demonstrate that ExpertGenQA achieves twice the efficiency of baseline few-shot approaches while maintaining $94.4\%$ topic coverage. Through systematic evaluation, we show that current LLM-based judges and reward models exhibit strong bias toward superficial writing styles rather than content quality. Our analysis using Bloom's Taxonomy reveals that ExpertGenQA better preserves the cognitive complexity distribution of expert-written questions compared to template-based approaches. When used to train retrieval models, our generated queries improve top-1 accuracy by $13.02\%$ over baseline performance, demonstrating their effectiveness for downstream applications in technical domains.

Attention Eclipse: Manipulating Attention to Bypass LLM Safety-Alignment

Recent research has shown that carefully crafted jailbreak inputs can induce large language models to produce harmful outputs, despite safety measures such as alignment. It is important to anticipate the range of potential Jailbreak attacks to guide effective defenses and accurate assessment of model safety. In this paper, we present a new approach for generating highly effective Jailbreak attacks that manipulate the attention of the model to selectively strengthen or weaken attention among different parts of the prompt. By harnessing attention loss, we develop more effective jailbreak attacks, that are also transferrable. The attacks amplify the success rate of existing Jailbreak algorithms including GCG, AutoDAN, and ReNeLLM, while lowering their generation cost (for example, the amplified GCG attack achieves 91.2% ASR, vs. 67.9% for the original attack on Llama2-7B/AdvBench, using less than a third of the generation time).

Gender Bias Mitigation for Bangla Classification Tasks

In this study, we investigate gender bias in Bangla pretrained language models, a largely under explored area in low-resource languages. To assess this bias, we applied gender-name swapping techniques to existing datasets, creating four manually annotated, task-specific datasets for sentiment analysis, toxicity detection, hate speech detection, and sarcasm detection. By altering names and gender-specific terms, we ensured these datasets were suitable for detecting and mitigating gender bias. We then proposed a joint loss optimization technique to mitigate gender bias across task-specific pretrained models. Our approach was evaluated against existing bias mitigation methods, with results showing that our technique not only effectively reduces bias but also maintains competitive accuracy compared to other baseline approaches. To promote further research, we have made both our implementation and datasets publicly available https://github.com/sajib-kumar/Gender-Bias-Mitigation-From-Bangla-PLM

Unfair Alignment: Examining Safety Alignment Across Vision Encoder Layers in Vision-Language Models

Vision-language models (VLMs) have improved significantly in multi-modal tasks, but their more complex architecture makes their safety alignment more challenging than the alignment of large language models (LLMs). In this paper, we reveal an unfair distribution of safety across the layers of VLM's vision encoder, with earlier and middle layers being disproportionately vulnerable to malicious inputs compared to the more robust final layers. This 'cross-layer' vulnerability stems from the model's inability to generalize its safety training from the default architectural settings used during training to unseen or out-of-distribution scenarios, leaving certain layers exposed. We conduct a comprehensive analysis by projecting activations from various intermediate layers and demonstrate that these layers are more likely to generate harmful outputs when exposed to malicious inputs. Our experiments with LLaVA-1.5 and Llama 3.2 show discrepancies in attack success rates and toxicity scores across layers, indicating that current safety alignment strategies focused on a single default layer are insufficient.

Not All Heads Matter: A Head-Level KV Cache Compression Method with Integrated Retrieval and Reasoning

Key-Value (KV) caching is a common technique to enhance the computational efficiency of Large Language Models (LLMs), but its memory overhead grows rapidly with input length. Prior work has shown that not all tokens are equally important for text generation, proposing layer-level KV cache compression to selectively retain key information. Recognizing the distinct roles of attention heads in generation, we propose HeadKV, a head-level KV cache compression method, and HeadKV-R2, which leverages a novel contextual reasoning ability estimation for compression. Our approach operates at the level of individual heads, estimating their importance for contextual QA tasks that require both retrieval and reasoning capabilities. Extensive experiments across diverse benchmarks (LongBench, LooGLE), model architectures (e.g., Llama-3-8B-Instruct, Mistral-7B-Instruct), and long-context abilities tests demonstrate that our head-level KV cache compression significantly outperforms strong baselines, particularly in low-resource settings (KV size = 64 & 128). Notably, our method retains just 1.5% of the KV cache while achieving 97% of the performance of the full KV cache on the contextual question answering benchmark.

LaMMA-P: Generalizable Multi-Agent Long-Horizon Task Allocation and Planning with LM-Driven PDDL Planner

Language models (LMs) possess a strong capability to comprehend natural language, making them effective in translating human instructions into detailed plans for simple robot tasks. Nevertheless, it remains a significant challenge to handle long-horizon tasks, especially in subtask identification and allocation for cooperative heterogeneous robot teams. To address this issue, we propose a Language Model-Driven Multi-Agent PDDL Planner (LaMMA-P), a novel multi-agent task planning framework that achieves state-of-the-art performance on long-horizon tasks. LaMMA-P integrates the strengths of the LMs' reasoning capability and the traditional heuristic search planner to achieve a high success rate and efficiency while demonstrating strong generalization across tasks. Additionally, we create MAT-THOR, a comprehensive benchmark that features household tasks with two different levels of complexity based on the AI2-THOR environment. The experimental results demonstrate that LaMMA-P achieves a 105% higher success rate and 36% higher efficiency than existing LM-based multi-agent planners. The experimental videos, code, and datasets of this work as well as the detailed prompts used in each module are available at https://lamma-p.github.io.

Adversarial Attacks on Parts of Speech: An Empirical Study in Text-to-Image Generation

Recent studies show that text-to-image (T2I) models are vulnerable to adversarial attacks, especially with noun perturbations in text prompts. In this study, we investigate the impact of adversarial attacks on different POS tags within text prompts on the images generated by T2I models. We create a high-quality dataset for realistic POS tag token swapping and perform gradient-based attacks to find adversarial suffixes that mislead T2I models into generating images with altered tokens. Our empirical results show that the attack success rate (ASR) varies significantly among different POS tag categories, with nouns, proper nouns, and adjectives being the easiest to attack. We explore the mechanism behind the steering effect of adversarial suffixes, finding that the number of critical tokens and content fusion vary among POS tags, while features like suffix transferability are consistent across categories. We have made our implementation publicly available at - https://github.com/shahariar-shibli/Adversarial-Attack-on-POS-Tags.

How to Leverage Personal Textual Knowledge for Personalized Conversational Information Retrieval

Personalized conversational information retrieval (CIR) combines conversational and personalizable elements to satisfy various users' complex information needs through multi-turn interaction based on their backgrounds. The key promise is that the personal textual knowledge base (PTKB) can improve the CIR effectiveness because the retrieval results can be more related to the user's background. However, PTKB is noisy: not every piece of knowledge in PTKB is relevant to the specific query at hand. In this paper, we explore and test several ways to select knowledge from PTKB and use it for query reformulation by using a large language model (LLM). The experimental results show the PTKB might not always improve the search results when used alone, but LLM can help generate a more appropriate personalized query when high-quality guidance is provided.

TRAWL: Tensor Reduced and Approximated Weights for Large Language Models

Large language models (LLMs) have fundamentally transformed artificial intelligence, catalyzing recent advancements while imposing substantial environmental and computational burdens. We introduce TRAWL (Tensor Reduced and Approximated Weights for Large Language Models), a novel methodology for optimizing LLMs through tensor decomposition. TRAWL leverages diverse strategies to exploit matrices within transformer-based architectures, realizing notable performance enhancements without necessitating retraining. The most significant improvements were observed through a layer-by-layer intervention strategy, particularly when applied to fully connected weights of the final layers, yielding up to 16% enhancement in accuracy without the need for additional data or fine-tuning. These results underscore the importance of targeted and adaptive techniques in increasing the efficiency and effectiveness of large language model optimization, thereby promoting the development of more sustainable and accessible AI systems.