The rapid growth of Large Language Models (LLMs) has put forward the study of biases as a crucial field. It is important to assess the influence of different types of biases embedded in LLMs to ensure fair use in sensitive fields. Although there have been extensive works on bias assessment in English, such efforts are rare and scarce for a major language like Bangla. In this work, we examine two types of social biases in LLM generated outputs for Bangla language. Our main contributions in this work are: (1) bias studies on two different social biases for Bangla (2) a curated dataset for bias measurement benchmarking (3) two different probing techniques for bias detection in the context of Bangla. This is the first work of such kind involving bias assessment of LLMs for Bangla to the best of our knowledge. All our code and resources are publicly available for the progress of bias related research in Bangla NLP.

The overwhelming success of GPT-4 in early 2023 highlighted the transformative potential of large language models (LLMs) across various sectors, including national security. This article explores the implications of LLM integration within national security contexts, analyzing their potential to revolutionize information processing, decision-making, and operational efficiency. Whereas LLMs offer substantial benefits, such as automating tasks and enhancing data analysis, they also pose significant risks, including hallucinations, data privacy concerns, and vulnerability to adversarial attacks. Through their coupling with decision-theoretic principles and Bayesian reasoning, LLMs can significantly improve decision-making processes within national security organizations. Namely, LLMs can facilitate the transition from data to actionable decisions, enabling decision-makers to quickly receive and distill available information with less manpower. Current applications within the US Department of Defense and beyond are explored, e.g., the USAF's use of LLMs for wargaming and automatic summarization, that illustrate their potential to streamline operations and support decision-making. However, these applications necessitate rigorous safeguards to ensure accuracy and reliability. The broader implications of LLM integration extend to strategic planning, international relations, and the broader geopolitical landscape, with adversarial nations leveraging LLMs for disinformation and cyber operations, emphasizing the need for robust countermeasures. Despite exhibiting "sparks" of artificial general intelligence, LLMs are best suited for supporting roles rather than leading strategic decisions. Their use in training and wargaming can provide valuable insights and personalized learning experiences for military personnel, thereby improving operational readiness.

Large Language Models (LLMs) have shown potential in reasoning over structured environments, e.g., knowledge graph and table. Such tasks typically require multi-hop reasoning, i.e., match natural language utterance with instances in the environment. Previous methods leverage LLMs to incrementally build a reasoning path, where the LLMs either invoke tools or pick up schemas by step-by-step interacting with the environment. We propose Reasoning-Path-Editing (Readi), a novel framework where LLMs can efficiently and faithfully reason over structured environments. In Readi, LLMs initially generate a reasoning path given a query, and edit the path only when necessary. We instantiate the path on structured environments and provide feedback to edit the path if anything goes wrong. Experimental results on three KGQA and two TableQA datasets show the effectiveness of Readi, significantly surpassing previous LLM-based methods (by 9.1% Hit@1 on WebQSP, 12.4% on MQA-3H and 9.5% on WTQ), comparable with state-of-the-art fine-tuned methods (67% on CWQ and 74.7% on WebQSP) and substantially boosting the vanilla LLMs (by 14.9% on CWQ). Our code will be available on https://aka.ms/readi.

When answering questions, LLMs can convey not only an answer, but a level of confidence about the answer being correct. This includes explicit confidence markers (e.g. giving a numeric score) as well as implicit markers, like an authoritative tone or elaborating with additional knowledge. For LLMs to be trustworthy knowledge sources, the confidence they convey should match their actual expertise; however, most current models tend towards overconfidence. To calibrate both implicit and explicit confidence markers, we introduce a pragmatic, listener-aware finetuning method (LACIE) that models the listener, considering not only whether an answer is right, but whether it will be accepted by a listener. We cast calibration as preference optimization, creating data via a two-agent game, where a speaker model's outputs are judged by a simulated listener. We then finetune three LLMs (Mistral-7B, Llama3-8B, Llama3-70B) with LACIE, and show that the resulting models are better calibrated w.r.t. a simulated listener. Crucially, these trends transfer to human listeners, helping them correctly predict model correctness: we conduct a human evaluation where annotators accept or reject an LLM's answers, finding that training with LACIE results in 47% fewer incorrect answers being accepted while maintaining the same level of acceptance for correct answers. Furthermore, LACIE generalizes to another dataset, resulting in a large increase in truthfulness on TruthfulQA when trained on TriviaQA. Our analysis indicates that LACIE leads to a better confidence separation between correct and incorrect examples. Qualitatively, we find that a LACIE-trained model hedges more and implicitly signals certainty when it is correct by using an authoritative tone or including details. Finally, LACIE finetuning leads to an emergent increase in model abstention (e.g. saying "I don't know") for answers that are likely wrong.

In recent years, Large Language Models (LLMs) have demonstrated remarkable capabilities in understanding and solving mathematical problems, leading to advancements in various fields. We propose an LLM-embodied path planning framework for mobile agents, focusing on solving high-level coverage path planning issues and low-level control. Our proposed multi-layer architecture uses prompted LLMs in the path planning phase and integrates them with the mobile agents' low-level actuators. To evaluate the performance of various LLMs, we propose a coverage-weighted path planning metric to assess the performance of the embodied models. Our experiments show that the proposed framework improves LLMs' spatial inference abilities. We demonstrate that the proposed multi-layer framework significantly enhances the efficiency and accuracy of these tasks by leveraging the natural language understanding and generative capabilities of LLMs. Our experiments show that this framework can improve LLMs' 2D plane reasoning abilities and complete coverage path planning tasks. We also tested three LLM kernels: gpt-4o, gemini-1.5-flash, and claude-3.5-sonnet. The experimental results show that claude-3.5 can complete the coverage planning task in different scenarios, and its indicators are better than those of the other models.

When LLMs are deployed in sensitive, human-facing settings, it is crucial that they do not output unsafe, biased, or privacy-violating outputs. For this reason, models are both trained and instructed to refuse to answer unsafe prompts such as "Tell me how to build a bomb." We find that, despite these safeguards, it is possible to break model defenses simply by appending a space to the end of a model's input. In a study of eight open-source models, we demonstrate that this acts as a strong enough attack to cause the majority of models to generate harmful outputs with very high success rates. We examine the causes of this behavior, finding that the contexts in which single spaces occur in tokenized training data encourage models to generate lists when prompted, overriding training signals to refuse to answer unsafe requests. Our findings underscore the fragile state of current model alignment and promote the importance of developing more robust alignment methods. Code and data will be available at https://github.com/hannah-aught/space_attack.

Automatic readability assessment is relevant to building NLP applications for education, content analysis, and accessibility. However, Arabic readability assessment is a challenging task due to Arabic's morphological richness and limited readability resources. In this paper, we present a set of experimental results on Arabic readability assessment using a diverse range of approaches, from rule-based methods to Arabic pretrained language models. We report our results on a newly created corpus at different textual granularity levels (words and sentence fragments). Our results show that combining different techniques yields the best results, achieving an overall macro F1 score of 86.7 at the word level and 87.9 at the fragment level on a blind test set. We make our code, data, and pretrained models publicly available.

The flourishing ecosystem centered around voice personal assistants (VPA), such as Amazon Alexa, has led to the booming of VPA apps. The largest app market Amazon skills store, for example, hosts over 200,000 apps. Despite their popularity, the open nature of app release and the easy accessibility of apps also raise significant concerns regarding security, privacy and quality. Consequently, various testing approaches have been proposed to systematically examine VPA app behaviors. To tackle the inherent lack of a visible user interface in the VPA app, two strategies are employed during testing, i.e., chatbot-style testing and model-based testing. The former often lacks effective guidance for expanding its search space, while the latter falls short in interpreting the semantics of conversations to construct precise and comprehensive behavior models for apps. In this work, we introduce Elevate, a model-enhanced large language model (LLM)-driven VUI testing framework. Elevate leverages LLMs' strong capability in natural language processing to compensate for semantic information loss during model-based VUI testing. It operates by prompting LLMs to extract states from VPA apps' outputs and generate context-related inputs. During the automatic interactions with the app, it incrementally constructs the behavior model, which facilitates the LLM in generating inputs that are highly likely to discover new states. Elevate bridges the LLM and the behavior model with innovative techniques such as encoding behavior model into prompts and selecting LLM-generated inputs based on the context relevance. Elevate is benchmarked on 4,000 real-world Alexa skills, against the state-of-the-art tester Vitas. It achieves 15% higher state space coverage compared to Vitas on all types of apps, and exhibits significant advancement in efficiency.

The massive amounts of web-mined parallel data contain large amounts of noise. Semantic misalignment, as the primary source of the noise, poses a challenge for training machine translation systems. In this paper, we first study the impact of real-world hard-to-detect misalignment noise by proposing a process to simulate the realistic misalignment controlled by semantic similarity. After quantitatively analyzing the impact of simulated misalignment on machine translation, we show the limited effectiveness of widely used pre-filters to improve the translation performance, underscoring the necessity of more fine-grained ways to handle data noise. By observing the increasing reliability of the model's self-knowledge for distinguishing misaligned and clean data at the token-level, we propose a self-correction approach which leverages the model's prediction distribution to revise the training supervision from the ground-truth data over training time. Through comprehensive experiments, we show that our self-correction method not only improves translation performance in the presence of simulated misalignment noise but also proves effective for real-world noisy web-mined datasets across eight translation tasks.

In this paper, we introduce the second-place solution in the KDD-2024 OAG-Challenge paper source tracing track. Our solution is mainly based on two methods, BERT and GCN, and combines the reasoning results of BERT and GCN in the final submission to achieve complementary performance. In the BERT solution, we focus on processing the fragments that appear in the references of the paper, and use a variety of operations to reduce the redundant interference in the fragments, so that the information received by BERT is more refined. In the GCN solution, we map information such as paper fragments, abstracts, and titles to a high-dimensional semantic space through an embedding model, and try to build edges between titles, abstracts, and fragments to integrate contextual relationships for judgment. In the end, our solution achieved a remarkable score of 0.47691 in the competition.