Evaluating Mathematical Reasoning Beyond Accuracy

The leaderboard of Large Language Models (LLMs) in mathematical tasks has been continuously updated. However, the majority of evaluations focus solely on the final results, neglecting the quality of the intermediate steps. This oversight can mask underlying problems, such as logical errors or unnecessary steps in the reasoning process. To measure reasoning beyond final-answer accuracy, we introduce ReasonEval, a new methodology for evaluating the quality of reasoning steps. ReasonEval employs $\textit{validity}$ and $\textit{redundancy}$ to characterize the reasoning quality, as well as accompanying LLMs to assess them automatically. Instantiated by base models that possess strong mathematical knowledge and trained with high-quality labeled data, ReasonEval achieves state-of-the-art performance on human-labeled datasets and can accurately detect different types of errors generated by perturbation. When applied to evaluate LLMs specialized in math, we find that an increase in final-answer accuracy does not necessarily guarantee an improvement in the overall quality of the reasoning steps for challenging mathematical problems. Additionally, we observe that ReasonEval can play a significant role in data selection. We release the best-performing model, meta-evaluation script, and all evaluation results at https://github.com/GAIR-NLP/ReasonEval.

On the Role of Summary Content Units in Text Summarization Evaluation

At the heart of the Pyramid evaluation method for text summarization lie human written summary content units (SCUs). These SCUs are concise sentences that decompose a summary into small facts. Such SCUs can be used to judge the quality of a candidate summary, possibly partially automated via natural language inference (NLI) systems. Interestingly, with the aim to fully automate the Pyramid evaluation, Zhang and Bansal (2021) show that SCUs can be approximated by automatically generated semantic role triplets (STUs). However, several questions currently lack answers, in particular: i) Are there other ways of approximating SCUs that can offer advantages? ii) Under which conditions are SCUs (or their approximations) offering the most value? In this work, we examine two novel strategies to approximate SCUs: generating SCU approximations from AMR meaning representations (SMUs) and from large language models (SGUs), respectively. We find that while STUs and SMUs are competitive, the best approximation quality is achieved by SGUs. We also show through a simple sentence-decomposition baseline (SSUs) that SCUs (and their approximations) offer the most value when ranking short summaries, but may not help as much when ranking systems or longer summaries.

Medical Unlearnable Examples: Securing Medical Data from Unauthorized Traning via Sparsity-Aware Local Masking

With the rapid growth of artificial intelligence (AI) in healthcare, there has been a significant increase in the generation and storage of sensitive medical data. This abundance of data, in turn, has propelled the advancement of medical AI technologies. However, concerns about unauthorized data exploitation, such as training commercial AI models, often deter researchers from making their invaluable datasets publicly available. In response to the need to protect this hard-to-collect data while still encouraging medical institutions to share it, one promising solution is to introduce imperceptible noise into the data. This method aims to safeguard the data against unauthorized training by inducing degradation in model generalization. Although existing methods have shown commendable data protection capabilities in general domains, they tend to fall short when applied to biomedical data, mainly due to their failure to account for the sparse nature of medical images. To address this problem, we propose the Sparsity-Aware Local Masking (SALM) method, a novel approach that selectively perturbs significant pixel regions rather than the entire image as previous strategies have done. This simple-yet-effective approach significantly reduces the perturbation search space by concentrating on local regions, thereby improving both the efficiency and effectiveness of data protection for biomedical datasets characterized by sparse features. Besides, we have demonstrated that SALM maintains the essential characteristics of the data, ensuring its clinical utility remains uncompromised. Our extensive experiments across various datasets and model architectures demonstrate that SALM effectively prevents unauthorized training of deep-learning models and outperforms previous state-of-the-art data protection methods.

Sora: A Review on Background, Technology, Limitations, and Opportunities of Large Vision Models

Sora is a text-to-video generative AI model, released by OpenAI in February 2024. The model is trained to generate videos of realistic or imaginative scenes from text instructions and show potential in simulating the physical world. Based on public technical reports and reverse engineering, this paper presents a comprehensive review of the model's background, related technologies, applications, remaining challenges, and future directions of text-to-video AI models. We first trace Sora's development and investigate the underlying technologies used to build this "world simulator". Then, we describe in detail the applications and potential impact of Sora in multiple industries ranging from film-making and education to marketing. We discuss the main challenges and limitations that need to be addressed to widely deploy Sora, such as ensuring safe and unbiased video generation. Lastly, we discuss the future development of Sora and video generation models in general, and how advancements in the field could enable new ways of human-AI interaction, boosting productivity and creativity of video generation.

Calibrating Long-form Generations from Large Language Models

To enhance Large Language Models' (LLMs) reliability, calibration is essential -- the model's assessed confidence scores should align with the actual likelihood of its responses being correct. However, current confidence elicitation methods and calibration metrics typically rely on a binary true/false assessment of response correctness. This approach does not apply to long-form generation, where an answer can be partially correct. Addressing this gap, we introduce a unified calibration framework, in which both the correctness of the LLMs' responses and their associated confidence levels are treated as distributions across a range of scores. Within this framework, we develop three metrics to precisely evaluate LLM calibration and further propose two confidence elicitation methods based on self-consistency and self-evaluation. Our experiments, which include long-form QA and summarization tasks, demonstrate that larger models don't necessarily guarantee better calibration, that calibration performance is found to be metric-dependent, and that self-consistency methods excel in factoid datasets. We also find that calibration can be enhanced through techniques such as fine-tuning, integrating relevant source documents, scaling the temperature, and combining self-consistency with self-evaluation. Lastly, we showcase a practical application of our system: selecting and cascading open-source models and ChatGPT to optimize correctness given a limited API budget. This research not only challenges existing notions of LLM calibration but also offers practical methodologies for improving trustworthiness in long-form generation.

TrustLLM: Trustworthiness in Large Language Models

Large language models (LLMs), exemplified by ChatGPT, have gained considerable attention for their excellent natural language processing capabilities. Nonetheless, these LLMs present many challenges, particularly in the realm of trustworthiness. Therefore, ensuring the trustworthiness of LLMs emerges as an important topic. This paper introduces TrustLLM, a comprehensive study of trustworthiness in LLMs, including principles for different dimensions of trustworthiness, established benchmark, evaluation, and analysis of trustworthiness for mainstream LLMs, and discussion of open challenges and future directions. Specifically, we first propose a set of principles for trustworthy LLMs that span eight different dimensions. Based on these principles, we further establish a benchmark across six dimensions including truthfulness, safety, fairness, robustness, privacy, and machine ethics. We then present a study evaluating 16 mainstream LLMs in TrustLLM, consisting of over 30 datasets. Our findings firstly show that in general trustworthiness and utility (i.e., functional effectiveness) are positively related. Secondly, our observations reveal that proprietary LLMs generally outperform most open-source counterparts in terms of trustworthiness, raising concerns about the potential risks of widely accessible open-source LLMs. However, a few open-source LLMs come very close to proprietary ones. Thirdly, it is important to note that some LLMs may be overly calibrated towards exhibiting trustworthiness, to the extent that they compromise their utility by mistakenly treating benign prompts as harmful and consequently not responding. Finally, we emphasize the importance of ensuring transparency not only in the models themselves but also in the technologies that underpin trustworthiness. Knowing the specific trustworthy technologies that have been employed is crucial for analyzing their effectiveness.

GOODAT: Towards Test-time Graph Out-of-Distribution Detection

Graph neural networks (GNNs) have found widespread application in modeling graph data across diverse domains. While GNNs excel in scenarios where the testing data shares the distribution of their training counterparts (in distribution, ID), they often exhibit incorrect predictions when confronted with samples from an unfamiliar distribution (out-of-distribution, OOD). To identify and reject OOD samples with GNNs, recent studies have explored graph OOD detection, often focusing on training a specific model or modifying the data on top of a well-trained GNN. Despite their effectiveness, these methods come with heavy training resources and costs, as they need to optimize the GNN-based models on training data. Moreover, their reliance on modifying the original GNNs and accessing training data further restricts their universality. To this end, this paper introduces a method to detect Graph Out-of-Distribution At Test-time (namely GOODAT), a data-centric, unsupervised, and plug-and-play solution that operates independently of training data and modifications of GNN architecture. With a lightweight graph masker, GOODAT can learn informative subgraphs from test samples, enabling the capture of distinct graph patterns between OOD and ID samples. To optimize the graph masker, we meticulously design three unsupervised objective functions based on the graph information bottleneck principle, motivating the masker to capture compact yet informative subgraphs for OOD detection. Comprehensive evaluations confirm that our GOODAT method outperforms state-of-the-art benchmarks across a variety of real-world datasets. The code is available at Github: https://github.com/Ee1s/GOODAT

Improving Faithfulness for Vision Transformers

Vision Transformers (ViTs) have achieved state-of-the-art performance for various vision tasks. One reason behind the success lies in their ability to provide plausible innate explanations for the behavior of neural architectures. However, ViTs suffer from issues with explanation faithfulness, as their focal points are fragile to adversarial attacks and can be easily changed with even slight perturbations on the input image. In this paper, we propose a rigorous approach to mitigate these issues by introducing Faithful ViTs (FViTs). Briefly speaking, an FViT should have the following two properties: (1) The top-$k$ indices of its self-attention vector should remain mostly unchanged under input perturbation, indicating stable explanations; (2) The prediction distribution should be robust to perturbations. To achieve this, we propose a new method called Denoised Diffusion Smoothing (DDS), which adopts randomized smoothing and diffusion-based denoising. We theoretically prove that processing ViTs directly with DDS can turn them into FViTs. We also show that Gaussian noise is nearly optimal for both $\ell_2$ and $\ell_\infty$-norm cases. Finally, we demonstrate the effectiveness of our approach through comprehensive experiments and evaluations. Specifically, we compare our FViTs with other baselines through visual interpretation and robustness accuracy under adversarial attacks. Results show that FViTs are more robust against adversarial attacks while maintaining the explainability of attention, indicating higher faithfulness.

Vector-Quantized Prompt Learning for Paraphrase Generation

Deep generative modeling of natural languages has achieved many successes, such as producing fluent sentences and translating from one language into another. However, the development of generative modeling techniques for paraphrase generation still lags behind largely due to the challenges in addressing the complex conflicts between expression diversity and semantic preservation. This paper proposes to generate diverse and high-quality paraphrases by exploiting the pre-trained models with instance-dependent prompts. To learn generalizable prompts, we assume that the number of abstract transforming patterns of paraphrase generation (governed by prompts) is finite and usually not large. Therefore, we present vector-quantized prompts as the cues to control the generation of pre-trained models. Extensive experiments demonstrate that the proposed method achieves new state-of-art results on three benchmark datasets, including Quora, Wikianswers, and MSCOCO. We will release all the code upon acceptance.