Recently, there has been a growing trend of utilizing Large Language Model (LLM) to evaluate the quality of other LLMs. Many studies have employed proprietary close-source models, especially GPT-4, as the evaluator. Alternatively, other works have fine-tuned judge models based on open-source LLMs as the evaluator. While the fine-tuned judge models are claimed to achieve comparable evaluation capability with GPT-4, in this study, we conduct an empirical study of judge models. Our findings indicate that although the fine-tuned judge models achieve high performance on in-domain test sets, even surpassing GPT-4, they underperform GPT-4 across several dimensions, including generalizability, fairness, aspect-specific evaluation, and scalability. We also reveal that the fine-tuned judge model inherently operates as a task-specific classifier, consequently imposing the limitations. Finally, we propose an effective indicator to measure the reliability of fine-tuned judges, with the aim of maximizing their utility in LLM evaluation.

Analyzing time series of fluxes from stars, known as stellar light curves, can reveal valuable information about stellar properties. However, most current methods rely on extracting summary statistics, and studies using deep learning have been limited to supervised approaches. In this research, we investigate the scaling law properties that emerge when learning from astronomical time series data using self-supervised techniques. By employing the GPT-2 architecture, we show the learned representation improves as the number of parameters increases from $10^4$ to $10^9$, with no signs of performance plateauing. We demonstrate that a self-supervised Transformer model achieves 3-10 times the sample efficiency compared to the state-of-the-art supervised learning model when inferring the surface gravity of stars as a downstream task. Our research lays the groundwork for analyzing stellar light curves by examining them through large-scale auto-regressive generative models.

Large Language Models (LLMs) are trained on extensive web corpora, which enable them to understand and generate human-like text. However, this training process also results in inherent biases within the models. These biases arise from web data's diverse and often uncurated nature, containing various stereotypes and prejudices. Previous works on debiasing models rely on benchmark datasets to measure their method's performance. However, these datasets suffer from several pitfalls due to the highly subjective understanding of bias, highlighting a critical need for contextual exploration. We propose understanding the context of inputs by considering the diverse situations in which they may arise. Our contribution is two-fold: (i) we augment 2,291 stereotyped statements from two existing bias-benchmark datasets with points for adding context; (ii) we develop the Context-Oriented Bias Indicator and Assessment Score (COBIAS) to assess a statement's contextual reliability in measuring bias. Our metric aligns with human judgment on contextual reliability of statements (Spearman's $\rho = 0.65, p = 3.4 * 10^{-60}$) and can be used to create reliable datasets, which would assist bias mitigation works.

We propose PLAYER*, a novel framework that addresses the limitations of existing agent-based approaches built on Large Language Models (LLMs) in handling complex questions and understanding interpersonal relationships in dynamic environments. PLAYER* enhances path planning in Murder Mystery Games (MMGs) using an anytime sampling-based planner and a questioning-driven search framework. By equipping agents with a set of sensors, PLAYER* eliminates the need for pre-defined questions and enables agents to navigate complex social interactions. We additionally make a contribution by introducing a quantifiable evaluation method using multiple-choice questions and present WellPlay, a dataset containing 1,482 question-answer pairs. Experimental results demonstrate PLAYER*'s superiority over existing multi-agent methods, enhancing the generalisability and adaptability of agents in MMGs and paving the way for more effective multi-agent interactions.

Large-scale pre-trained generative models are taking the world by storm, due to their abilities in generating creative content. Meanwhile, safeguards for these generative models are developed, to protect users' rights and safety, most of which are designed for large language models. Existing methods primarily focus on jailbreak and adversarial attacks, which mainly evaluate the model's safety under malicious prompts. Recent work found that manually crafted safe prompts can unintentionally trigger unsafe generations. To further systematically evaluate the safety risks of text-to-image models, we propose a novel Automatic Red-Teaming framework, ART. Our method leverages both vision language model and large language model to establish a connection between unsafe generations and their prompts, thereby more efficiently identifying the model's vulnerabilities. With our comprehensive experiments, we reveal the toxicity of the popular open-source text-to-image models. The experiments also validate the effectiveness, adaptability, and great diversity of ART. Additionally, we introduce three large-scale red-teaming datasets for studying the safety risks associated with text-to-image models. Datasets and models can be found in https://github.com/GuanlinLee/ART.

The rapid advancement of Large Language Models (LLMs) has introduced a new frontier in natural language processing, particularly in understanding and processing long-context information. However, the evaluation of these models' long-context abilities remains a challenge due to the limitations of current benchmarks. To address this gap, we introduce NovelQA, a benchmark specifically designed to test the capabilities of LLMs with extended texts. Constructed from English novels, NovelQA offers a unique blend of complexity, length, and narrative coherence, making it an ideal tool for assessing deep textual understanding in LLMs. This paper presents the design and construction of NovelQA, highlighting its manual annotation, and diverse question types. Our evaluation of Long-context LLMs on NovelQA reveals significant insights into the models' performance, particularly emphasizing the challenges they face with multi-hop reasoning, detail-oriented questions, and extremely long input with an average length more than 200,000 tokens. The results underscore the necessity for further advancements in LLMs to improve their long-context comprehension.

As world knowledge evolves and new task paradigms emerge, Continual Learning (CL) is crucial for keeping Large Language Models (LLMs) up-to-date and addressing their shortcomings. In practical applications, LLMs often require both continual instruction tuning (CIT) and continual pre-training (CPT) to adapt to new task paradigms and acquire necessary knowledge for task-solving. However, it remains challenging to collect CPT data that addresses the knowledge deficiencies in models while maintaining adequate volume, and improving the efficiency of utilizing this data also presents significant difficulties. Inspired by the 'summarizing mistakes' learning skill, we propose the Continue Evolving from Mistakes (CEM) method, aiming to provide a data-efficient approach for collecting CPT data and continually improving LLMs' performance through iterative evaluation and supplementation with mistake-relevant knowledge. To efficiently utilize these CPT data and mitigate forgetting, we design a novel CL training set construction paradigm that integrates parallel CIT and CPT data. Extensive experiments demonstrate the efficacy of the CEM method, achieving up to a 17% improvement in accuracy in the best case. Furthermore, additional experiments confirm the potential of combining CEM with catastrophic forgetting mitigation methods, enabling iterative and continual model evolution.

Text watermarks for large language models (LLMs) have been commonly used to identify the origins of machine-generated content, which is promising for assessing liability when combating deepfake or harmful content. While existing watermarking techniques typically prioritize robustness against removal attacks, unfortunately, they are vulnerable to spoofing attacks: malicious actors can subtly alter the meanings of LLM-generated responses or even forge harmful content, potentially misattributing blame to the LLM developer. To overcome this, we introduce a bi-level signature scheme, Bileve, which embeds fine-grained signature bits for integrity checks (mitigating spoofing attacks) as well as a coarse-grained signal to trace text sources when the signature is invalid (enhancing detectability) via a novel rank-based sampling strategy. Compared to conventional watermark detectors that only output binary results, Bileve can differentiate 5 scenarios during detection, reliably tracing text provenance and regulating LLMs. The experiments conducted on OPT-1.3B and LLaMA-7B demonstrate the effectiveness of Bileve in defeating spoofing attacks with enhanced detectability. Warning: This paper contains examples of offensive language due to attacks.

Recent advancements in general-purpose AI have highlighted the importance of guiding AI systems towards the intended goals, ethical principles, and values of individuals and groups, a concept broadly recognized as alignment. However, the lack of clarified definitions and scopes of human-AI alignment poses a significant obstacle, hampering collaborative efforts across research domains to achieve this alignment. In particular, ML- and philosophy-oriented alignment research often views AI alignment as a static, unidirectional process (i.e., aiming to ensure that AI systems' objectives match humans) rather than an ongoing, mutual alignment problem [429]. This perspective largely neglects the long-term interaction and dynamic changes of alignment. To understand these gaps, we introduce a systematic review of over 400 papers published between 2019 and January 2024, spanning multiple domains such as Human-Computer Interaction (HCI), Natural Language Processing (NLP), Machine Learning (ML), and others. We characterize, define and scope human-AI alignment. From this, we present a conceptual framework of "Bidirectional Human-AI Alignment" to organize the literature from a human-centered perspective. This framework encompasses both 1) conventional studies of aligning AI to humans that ensures AI produces the intended outcomes determined by humans, and 2) a proposed concept of aligning humans to AI, which aims to help individuals and society adjust to AI advancements both cognitively and behaviorally. Additionally, we articulate the key findings derived from literature analysis, including discussions about human values, interaction techniques, and evaluations. To pave the way for future studies, we envision three key challenges for future directions and propose examples of potential future solutions.

Self-correction has achieved impressive results in enhancing the style and security of the generated output from large language models (LLMs). However, recent studies suggest that self-correction might be limited or even counterproductive in reasoning tasks due to LLMs' difficulties in identifying logical mistakes. In this paper, we aim to enhance the self-checking capabilities of LLMs by constructing training data for checking tasks. Specifically, we apply the Chain of Thought (CoT) methodology to self-checking tasks, utilizing fine-grained step-level analyses and explanations to assess the correctness of reasoning paths. We propose a specialized checking format called "Step CoT Check". Following this format, we construct a checking-correction dataset that includes detailed step-by-step analysis and checking. Then we fine-tune LLMs to enhance their error detection and correction abilities. Our experiments demonstrate that fine-tuning with the "Step CoT Check" format significantly improves the self-checking and self-correction abilities of LLMs across multiple benchmarks. This approach outperforms other formats, especially in locating the incorrect position, with greater benefits observed in larger models. For reproducibility, all the datasets and code are provided in https://github.com/bammt/Learn-to-check.