[Notice: This version is outdated. Recent research contradicts some key claims; we are working on a major revision with more nuanced analysis. Please wait for the updated version.] While LLMs excel at multi-hop questions (e.g. "Who is the spouse of the performer of Imagine?") when using chain-of-thought reasoning (CoT), they struggle when forced to reason internally (without CoT). Previous work on the size and nature of this gap produced mixed evidence with inconclusive results. In this paper, we introduce a controlled setting for investigating two-hop reasoning in LLMs, where the above-chance performance constitutes undeniable evidence for latent reasoning. We fine-tune LLMs (including Llama 3 8B Instruct and GPT-4o) on fictional facts and confirm that they generalize to answering two-hop questions about them using CoT. We find that models can perform latent reasoning when facts appear together during training or in the prompt. However, to our surprise, models completely fail at two-hop reasoning without CoT when learned facts only appear in different documents, achieving chance-level accuracy and chance-level test loss. We call this complete failure to compose separately learned facts the Two-Hop Curse. Moreover, we evaluate 9 frontier LLMs on real-world facts, finding that models completely fail at two-hop no-CoT reasoning for over half of question categories while maintaining partial success with CoT across most categories. These results suggest that LLMs lack a general capability for latent multi-hop reasoning independent of the question type.

In this study, we explore the use of a transformer-based language model as an encoder to predict the band gaps of semiconductor materials directly from their text descriptions. Quantum chemistry simulations, including Density Functional Theory (DFT), are computationally intensive and time-consuming, which limits their practicality for high-throughput material screening, particularly for complex systems. Shallow machine learning (ML) models, while effective, often require extensive data preprocessing to convert non-numerical material properties into numerical inputs. In contrast, our approach leverages textual data directly, bypassing the need for complex feature engineering. We generate material descriptions in two formats: formatted strings combining features and natural language text generated using the ChatGPT API. We demonstrate that the RoBERTa model, pre-trained on natural language processing tasks, performs effectively as an encoder for prediction tasks. With minimal fine-tuning, it achieves a mean absolute error (MAE) of approximately 0.33 eV, performing better than shallow machine learning models such as Support Vector Regression, Random Forest, and XGBoost. Even when only the linear regression head is trained while keeping the RoBERTa encoder layers frozen, the accuracy remains nearly identical to that of the fully trained model. This demonstrates that the pre-trained RoBERTa encoder is highly adaptable for processing domain-specific text related to material properties, such as the band gap, significantly reducing the need for extensive retraining. This study highlights the potential of transformer-based language models to serve as efficient and versatile encoders for semiconductor materials property prediction tasks.

Honorifics serve as powerful linguistic markers that reflect social hierarchies and cultural values. This paper presents a large-scale, cross-linguistic exploration of usage of honorific pronouns in Bengali and Hindi Wikipedia articles, shedding light on how socio-cultural factors shape language. Using LLM (GPT-4o), we annotated 10, 000 articles of real and fictional beings in each language for several sociodemographic features such as gender, age, fame, and exoticness, and the use of honorifics. We find that across all feature combinations, use of honorifics is consistently more common in Bengali than Hindi. For both languages, the use non-honorific pronouns is more commonly observed for infamous, juvenile, and exotic beings. Notably, we observe a gender bias in use of honorifics in Hindi, with men being more commonly referred to with honorifics than women.

Large Language Models (LLMs) have been observed to exhibit bias in numerous ways, potentially creating or worsening outcomes for specific groups identified by protected attributes such as sex, race, sexual orientation, or age. To help address this gap, we introduce LangFair, an open-source Python package that aims to equip LLM practitioners with the tools to evaluate bias and fairness risks relevant to their specific use cases. The package offers functionality to easily generate evaluation datasets, comprised of LLM responses to use-case-specific prompts, and subsequently calculate applicable metrics for the practitioner's use case. To guide in metric selection, LangFair offers an actionable decision framework.

The crisis of mental health issues is escalating. Effective counseling serves as a critical lifeline for individuals suffering from conditions like PTSD, stress, etc. Therapists forge a crucial therapeutic bond with clients, steering them towards positivity. Unfortunately, the massive shortage of professionals, high costs, and mental health stigma pose significant barriers to consulting therapists. As a substitute, Virtual Mental Health Assistants (VMHAs) have emerged in the digital healthcare space. However, most existing VMHAs lack the commonsense to understand the nuanced sentiments of clients to generate effective responses. To this end, we propose EmpRes, a novel sentiment-guided mechanism incorporating commonsense awareness for generating responses. By leveraging foundation models and harnessing commonsense knowledge, EmpRes aims to generate responses that effectively shape the client's sentiment towards positivity. We evaluate the performance of EmpRes on HOPE, a benchmark counseling dataset, and observe a remarkable performance improvement compared to the existing baselines across a suite of qualitative and quantitative metrics. Moreover, our extensive empirical analysis and human evaluation show that the generation ability of EmpRes is well-suited and, in some cases, surpasses the gold standard. Further, we deploy EmpRes as a chat interface for users seeking mental health support. We address the deployed system's effectiveness through an exhaustive user study with a significant positive response. Our findings show that 91% of users find the system effective, 80% express satisfaction, and over 85.45% convey a willingness to continue using the interface and recommend it to others, demonstrating the practical applicability of EmpRes in addressing the pressing challenges of mental health support, emphasizing user feedback, and ethical considerations in a real-world context.

We present a unified dataset for document Question-Answering (QA), which is obtained combining several public datasets related to Document AI and visually rich document understanding (VRDU). Our main contribution is twofold: on the one hand we reformulate existing Document AI tasks, such as Information Extraction (IE), into a Question-Answering task, making it a suitable resource for training and evaluating Large Language Models; on the other hand, we release the OCR of all the documents and include the exact position of the answer to be found in the document image as a bounding box. Using this dataset, we explore the impact of different prompting techniques (that might include bounding box information) on the performance of open-weight models, identifying the most effective approaches for document comprehension.

Visual-language models (VLM) have emerged as a powerful tool for learning a unified embedding space for vision and language. Inspired by large language models, which have demonstrated strong reasoning and multi-task capabilities, visual large language models (VLLMs) are gaining increasing attention for building general-purpose VLMs. Despite the significant progress made in VLLMs, the related literature remains limited, particularly from a comprehensive application perspective, encompassing generalized and specialized applications across vision (image, video, depth), action, and language modalities. In this survey, we focus on the diverse applications of VLLMs, examining their using scenarios, identifying ethics consideration and challenges, and discussing future directions for their development. By synthesizing these contents, we aim to provide a comprehensive guide that will pave the way for future innovations and broader applications of VLLMs. The paper list repository is available: https://github.com/JackYFL/awesome-VLLMs.

The chapter discusses the foundational impact of modern generative AI models on information access (IA) systems. In contrast to traditional AI, the large-scale training and superior data modeling of generative AI models enable them to produce high-quality, human-like responses, which brings brand new opportunities for the development of IA paradigms. In this chapter, we identify and introduce two of them in details, i.e., information generation and information synthesis. Information generation allows AI to create tailored content addressing user needs directly, enhancing user experience with immediate, relevant outputs. Information synthesis leverages the ability of generative AI to integrate and reorganize existing information, providing grounded responses and mitigating issues like model hallucination, which is particularly valuable in scenarios requiring precision and external knowledge. This chapter delves into the foundational aspects of generative models, including architecture, scaling, and training, and discusses their applications in multi-modal scenarios. Additionally, it examines the retrieval-augmented generation paradigm and other methods for corpus modeling and understanding, demonstrating how generative AI can enhance information access systems. It also summarizes potential challenges and fruitful directions for future studies.

Evaluating Large Language Models (LLMs) is crucial for understanding their capabilities and limitations across various applications, including natural language processing and code generation. Existing benchmarks like MMLU, C-Eval, and HumanEval assess general LLM performance but lack focus on specific expert domains such as cybersecurity. Previous attempts to create cybersecurity datasets have faced limitations, including insufficient data volume and a reliance on multiple-choice questions (MCQs). To address these gaps, we propose SecBench, a multi-dimensional benchmarking dataset designed to evaluate LLMs in the cybersecurity domain. SecBench includes questions in various formats (MCQs and short-answer questions (SAQs)), at different capability levels (Knowledge Retention and Logical Reasoning), in multiple languages (Chinese and English), and across various sub-domains. The dataset was constructed by collecting high-quality data from open sources and organizing a Cybersecurity Question Design Contest, resulting in 44,823 MCQs and 3,087 SAQs. Particularly, we used the powerful while cost-effective LLMs to (1). label the data and (2). constructing a grading agent for automatic evaluation of SAQs. Benchmarking results on 16 SOTA LLMs demonstrate the usability of SecBench, which is arguably the largest and most comprehensive benchmark dataset for LLMs in cybersecurity. More information about SecBench can be found at our website, and the dataset can be accessed via the artifact link.

We introduce FACTS Grounding, an online leaderboard and associated benchmark that evaluates language models' ability to generate text that is factually accurate with respect to given context in the user prompt. In our benchmark, each prompt includes a user request and a full document, with a maximum length of 32k tokens, requiring long-form responses. The long-form responses are required to be fully grounded in the provided context document while fulfilling the user request. Models are evaluated using automated judge models in two phases: (1) responses are disqualified if they do not fulfill the user request; (2) they are judged as accurate if the response is fully grounded in the provided document. The automated judge models were comprehensively evaluated against a held-out test-set to pick the best prompt template, and the final factuality score is an aggregate of multiple judge models to mitigate evaluation bias. The FACTS Grounding leaderboard will be actively maintained over time, and contains both public and private splits to allow for external participation while guarding the integrity of the leaderboard. It can be found at https://www.kaggle.com/facts-leaderboard.