Understanding and mitigating hallucinations in Large Language Models (LLMs) is crucial for ensuring reliable content generation. While previous research has primarily focused on "when" LLMs hallucinate, our work explains "why" and directly links model behaviour to the pre-training data that forms their prior knowledge. Specifically, we demonstrate that an asymmetry exists in the recognition of logically equivalent facts, which can be attributed to frequency discrepancies of entities appearing as subjects versus objects. Given that most pre-training datasets are inaccessible, we leverage the fully open-source OLMo series by indexing its Dolma dataset to estimate entity frequencies. Using relational facts (represented as triples) from Wikidata5M, we construct probing datasets to isolate this effect. Our experiments reveal that facts with a high-frequency subject and a low-frequency object are better recognised than their inverse, despite their logical equivalence. The pattern reverses in low-to-high frequency settings, and no statistically significant asymmetry emerges when both entities are high-frequency. These findings highlight the influential role of pre-training data in shaping model predictions and provide insights for inferring the characteristics of pre-training data in closed or partially closed LLMs.

Forecasting is an important task in many domains, such as technology and economics. However existing forecasting benchmarks largely lack comprehensive confidence assessment, focus on limited question types, and often consist of artificial questions that do not align with real-world human forecasting needs. To address these gaps, we introduce FOReCAst (Future Outcome Reasoning and Confidence Assessment), a benchmark that evaluates models' ability to make predictions and their confidence in them. FOReCAst spans diverse forecasting scenarios involving Boolean questions, timeframe prediction, and quantity estimation, enabling a comprehensive evaluation of both prediction accuracy and confidence calibration for real-world applications.

Recent works integrating Knowledge Graphs (KGs) have led to promising improvements in enhancing reasoning accuracy of Large Language Models (LLMs). However, current benchmarks focus mainly on closed-ended tasks, leaving a gap in the assessment of more complex real-world scenarios. This gap has also obscured the evaluation of KGs' potential to mitigate the problem of hallucination in LLMs. To fill the gap, we introduce OKGQA, a new benchmark specifically designed to assess LLMs enhanced with KGs under open-ended, real-world question answering scenarios. OKGQA is designed to closely reflect the complexities of practical applications using questions from different types, and incorporates specific metrics to measure both the reduction in hallucinations and the enhancement in reasoning capabilities. To consider the scenario in which KGs may have varying levels of mistakes, we propose another benchmark variant OKGQA-P to assess model performance when the semantics and structure of KGs are deliberately perturbed and contaminated. OKGQA aims to (1) explore whether KGs can make LLMs more trustworthy in an open-ended setting, and (2) conduct a comparative analysis to shed light on methods and future directions for leveraging KGs to reduce LLMs' hallucination. We believe that this study can facilitate a more complete performance comparison and encourage continuous improvement in integrating KGs with LLMs.

The Mixture-of-Experts (MoE) paradigm has emerged as a powerful approach for scaling transformers with improved resource utilization. However, efficiently fine-tuning MoE models remains largely underexplored. Inspired by recent works on Parameter-Efficient Fine-Tuning (PEFT), we present a unified framework for integrating PEFT modules directly into the MoE mechanism. Aligning with the core principles and architecture of MoE, our framework encompasses a set of design dimensions including various functional and composition strategies. By combining design choices within our framework, we introduce Parameter-Efficient Routed Fine-Tuning (PERFT) as a flexible and scalable family of PEFT strategies tailored for MoE models. Extensive experiments on adapting OLMoE-1B-7B and Mixtral-8$\times$7B for commonsense and arithmetic reasoning tasks demonstrate the effectiveness, scalability, and intriguing dynamics of PERFT. Additionally, we provide empirical findings for each specific design choice to facilitate better application of MoE and PEFT.

Current state-of-the-art dialogue systems heavily rely on extensive training datasets. However, challenges arise in domains where domain-specific training datasets are insufficient or entirely absent. To tackle this challenge, we propose a novel data \textbf{A}ugmentation framework for \textbf{M}ulti-\textbf{D}omain \textbf{D}ialogue \textbf{G}eneration, referred to as \textbf{AMD$^2$G}. The AMD$^2$G framework consists of a data augmentation process and a two-stage training approach: domain-agnostic training and domain adaptation training. We posit that domain corpora are a blend of domain-agnostic and domain-specific features, with certain representation patterns shared among diverse domains. Domain-agnostic training aims to enable models to learn these common expressive patterns. To construct domain-agnostic dialogue corpora, we employ a \textit{\textbf{de-domaining}} data processing technique used to remove domain-specific features. By mitigating the effects of domain-specific features, the model trained on the de-domained corpora can effectively learn common expression patterns in different domains. Subsequently, we adapt the learned domain-agnostic features to the target domain through domain adaptation training. We conduct experiments on Chinese dialogue datasets from five different domains and show that AMD$^2$G achieves superior performance compared to both direct training on the target domain corpus and collective training on all five domain corpora. Our work underscores AMD$^2$G as a viable alternative solution for low-resource multi-domain dialogue generation. Code and data associated with our work are available on GitHub repository$^{\text 1}$.

Various jailbreak attacks have been proposed to red-team Large Language Models (LLMs) and revealed the vulnerable safeguards of LLMs. Besides, some methods are not limited to the textual modality and extend the jailbreak attack to Multimodal Large Language Models (MLLMs) by perturbing the visual input. However, the absence of a universal evaluation benchmark complicates the performance reproduction and fair comparison. Besides, there is a lack of comprehensive evaluation of closed-source state-of-the-art (SOTA) models, especially MLLMs, such as GPT-4V. To address these issues, this work first builds a comprehensive jailbreak evaluation dataset with 1445 harmful questions covering 11 different safety policies. Based on this dataset, extensive red-teaming experiments are conducted on 11 different LLMs and MLLMs, including both SOTA proprietary models and open-source models. We then conduct a deep analysis of the evaluated results and find that (1) GPT4 and GPT-4V demonstrate better robustness against jailbreak attacks compared to open-source LLMs and MLLMs. Large Language Models (LLMs) and Multimodal Large Language Models (MLLMs) have shown superior performance in text generation. To avoid generating unobjectionable content learned from the web-scale training corpus, stringent safety regulations have been applied during the safety alignment (Ouyang et al., 2022; Touvron et al., 2023). However, many jailbreak attacks have been proven to be able to bypass these safeguards and successfully elicit harmful generations.

Recently, Multimodal LLMs (MLLMs) have shown a great ability to understand images. However, like traditional vision models, they are still vulnerable to adversarial images. Meanwhile, Chain-of-Thought (CoT) reasoning has been widely explored on MLLMs, which not only improves model's performance, but also enhances model's explainability by giving intermediate reasoning steps. Nevertheless, there is still a lack of study regarding MLLMs' adversarial robustness with CoT and an understanding of what the rationale looks like when MLLMs infer wrong answers with adversarial images. Our research evaluates the adversarial robustness of MLLMs when employing CoT reasoning, finding that CoT marginally improves adversarial robustness against existing attack methods. Moreover, we introduce a novel stop-reasoning attack technique that effectively bypasses the CoT-induced robustness enhancements. Finally, we demonstrate the alterations in CoT reasoning when MLLMs confront adversarial images, shedding light on their reasoning process under adversarial attacks.

In recent years, modeling evolving knowledge over temporal knowledge graphs (TKGs) has become a heated topic. Various methods have been proposed to forecast links on TKGs. Most of them are embedding-based, where hidden representations are learned to represent knowledge graph (KG) entities and relations based on the observed graph contexts. Although these methods show strong performance on traditional TKG forecasting (TKGF) benchmarks, they naturally face a strong challenge when they are asked to model the unseen zero-shot relations that has no prior graph context. In this paper, we try to mitigate this problem as follows. We first input the text descriptions of KG relations into large language models (LLMs) for generating relation representations, and then introduce them into embedding-based TKGF methods. LLM-empowered representations can capture the semantic information in the relation descriptions. This makes the relations, whether seen or unseen, with similar semantic meanings stay close in the embedding space, enabling TKGF models to recognize zero-shot relations even without any observed graph context. Experimental results show that our approach helps TKGF models to achieve much better performance in forecasting the facts with previously unseen relations, while still maintaining their ability in link forecasting regarding seen relations.

Question answering over temporal knowledge graphs (TKGQA) has recently found increasing interest. TKGQA requires temporal reasoning techniques to extract the relevant information from temporal knowledge bases. The only existing TKGQA dataset, i.e., CronQuestions, consists of temporal questions based on the facts from a fixed time period, where a temporal knowledge graph (TKG) spanning the same period can be fully used for answer inference, allowing the TKGQA models to use even the future knowledge to answer the questions based on the past facts. In real-world scenarios, however, it is also common that given the knowledge until now, we wish the TKGQA systems to answer the questions asking about the future. As humans constantly seek plans for the future, building TKGQA systems for answering such forecasting questions is important. Nevertheless, this has still been unexplored in previous research. In this paper, we propose a novel task: forecasting question answering over temporal knowledge graphs. We also propose a large-scale TKGQA benchmark dataset, i.e., ForecastTKGQuestions, for this task. It includes three types of questions, i.e., entity prediction, yes-no, and fact reasoning questions. For every forecasting question in our dataset, QA models can only have access to the TKG information before the timestamp annotated in the given question for answer inference. We find that the state-of-the-art TKGQA methods perform poorly on forecasting questions, and they are unable to answer yes-no questions and fact reasoning questions. To this end, we propose ForecastTKGQA, a TKGQA model that employs a TKG forecasting module for future inference, to answer all three types of questions. Experimental results show that ForecastTKGQA outperforms recent TKGQA methods on the entity prediction questions, and it also shows great effectiveness in answering the other two types of questions.

Stemming from traditional knowledge graphs (KGs), hyper-relational KGs (HKGs) provide additional key-value pairs (i.e., qualifiers) for each KG fact that help to better restrict the fact validity. In recent years, there has been an increasing interest in studying graph reasoning over HKGs. In the meantime, due to the ever-evolving nature of world knowledge, extensive parallel works have been focusing on reasoning over temporal KGs (TKGs), where each TKG fact can be viewed as a KG fact coupled with a timestamp (or time period) specifying its time validity. The existing HKG reasoning approaches do not consider temporal information because it is not explicitly specified in previous benchmark datasets. Besides, all the previous TKG reasoning methods only lay emphasis on temporal reasoning and have no way to learn from qualifiers. To this end, we aim to fill the gap between TKG reasoning and HKG reasoning. We develop two new benchmark hyper-relational TKG (HTKG) datasets, i.e., Wiki-hy and YAGO-hy, and propose a HTKG reasoning model that efficiently models both temporal facts and qualifiers. We further exploit additional time-invariant relational knowledge from the Wikidata knowledge base and study its effectiveness in HTKG reasoning. Time-invariant relational knowledge serves as the knowledge that remains unchanged in time (e.g., Sasha Obama is the child of Barack Obama), and it has never been fully explored in previous TKG reasoning benchmarks and approaches. Experimental results show that our model substantially outperforms previous related methods on HTKG link prediction and can be enhanced by jointly leveraging both temporal and time-invariant relational knowledge.