Safety concerns remain one of the most significant obstacles to the large-scale deployment and continued advancement of autonomous driving (AD) systems. A major underlying cause of many safety-related incidents in AD systems is suboptimal or erroneous decision-making when the vehicle encounters corner cases (CCs)—rare, unexpected, or extreme situations that fall outside typical operating conditions. Although recent advances in artificial intelligence have driven substantial progress in both autonomous driving and corner-case research, the field still lacks a coherent conceptual foundation and a systematic, widely accepted categorization of CCs. In this survey, we address this gap by offering a structured review of the existing corner-case literature along three key dimensions: understanding, detection, and generation. The key contribution is a three-level classification of corner cases—spanning data-level, model-level, and semantic-level CCs—that clarifies and disambiguates competing definitions and perspectives on AD corner cases. Building on this framework, we examine simulator-based methods for corner-case selection and data generation, and we identify open challenges, promising directions, and potential solutions to more effectively handle corner cases in autonomous driving systems.

Out-of-distribution (OOD) detection is crucial for deploying models in open-world environments. This process aims to mitigate the issue of overconfident predictions, which is a common problem for models designed for closed-domain tasks when they encounter OOD data. Recent progress in OOD detection has shown that integrating auxiliary datasets during model training can greatly enhance OOD detection performance. However, existing methods tend to heavily depend on these auxiliary datasets to establish the decision boundary for in-distribution (ID) data, while not adequately addressing OOD object detection in safety–critical applications. In this article, we propose object-level OOD detection and introduce a hierarchical hard negative sampling (HNS) strategy that does not necessitate auxiliary data. Specifically, we offer a new metric that strategically considers difficult negatives near the decision boundary between inter-class and intra-class instances. Inspired by adversarial thinking, we sample outliers for each class, ensuring that the negative samples capture both diversity and informative traits. We conducted comprehensive experiments on three public datasets. The results demonstrate that HNS performs superiorly in object-level OOD detection, even without auxiliary datasets. The source code can be accessed at: https://github.com/Aurevior1/HNS .

Recent advances such as OpenAI-o1 and DeepSeek R1 have demonstrated the potential of Reinforcement Learning (RL) to enhance reasoning abilities in Large Language Models (LLMs). While open-source replication efforts have primarily focused on mathematical and coding domains, methods and resources for developing general reasoning capabilities remain underexplored. This gap is partly due to the challenge of collecting diverse and verifiable reasoning data suitable for RL. We hypothesize that logical reasoning is critical for developing general reasoning capabilities, as logic forms a fundamental building block of reasoning. In this work, we present SynLogic, a data synthesis framework and dataset that generates diverse logical reasoning data at scale, encompassing 35 diverse logical reasoning tasks. The SynLogic approach enables controlled synthesis of data with adjustable difficulty and quantity. Importantly, all examples can be verified by simple rules, making them ideally suited for RL with verifiable rewards. In our experiments, we validate the effectiveness of RL training on the SynLogic dataset based on 7B and 32B models. SynLogic leads to state-of-the-art logical reasoning performance among open-source datasets, surpassing DeepSeek-R1-Distill-Qwen-32B by 6 points on BBEH. Furthermore, mixing SynLogic data with mathematical and coding tasks improves the training efficiency of these domains and significantly enhances reasoning generalization. Notably, our mixed training model outperforms DeepSeek-R1-Zero-Qwen-32B across multiple benchmarks. These findings position SynLogic as a valuable resource for advancing the broader reasoning capabilities of LLMs. We will open-source both the data synthesis pipeline and the SynLogic dataset.

While large language models (LLMs) have demonstrated remarkable abilities across various fields, hallucination remains a significant challenge.Recent studies have explored hallucinations through the lens of internal representations, proposing mechanisms to decipher LLMs' adherence to facts.However, these approaches often fail to generalize to out-of-distribution data, leading to concerns about whether internal representation patterns reflect fundamental factual awareness, or only overfit spurious correlations on the specific datasets.In this work, we investigate whether a universal truthfulness hyperplane that distinguishes the model's factually correct and incorrect outputs exists within the model.To this end, we scale up the number of training datasets and conduct an extensive evaluation -we train the truthfulness hyperplane on a diverse collection of over 40 datasets and examine its cross-task, cross-domain, and in-domain generalization.Our results indicate that increasing the diversity of the training datasets significantly enhances the performance in all scenarios, while the volume of data samples plays a less critical role.This finding supports the optimistic hypothesis that a universal truthfulness hyperplane may indeed exist within the model, offering promising directions for future research.Code is publicly available at https://github.com/hkust-nlp/ Universal_Truthfulness_Hyperplane.Tend to overfit

Large language models (LLMs) frequently hallucinate, e.g., making factual errors, yet our understanding of why they make these errors remains limited. In this study, we aim to understand the underlying mechanisms of LLM hallucinations from the perspective of *inner representations*. We discover a pattern associated with hallucinations: correct generations tend to have *sharper* context activations in the hidden states of the in-context tokens, compared to that of the incorrect generations. Leveraging this signal, we propose an entropy-based metric to quantify the *sharpness* among the in-context hidden states and incorporate it into the decoding process, i.e, use the entropy value to adjust the next token prediction distribution to improve the factuality and overall quality of the generated text. Experiments on knowledge-seeking datasets (Natural Questions, HotpotQA, TriviaQA) and hallucination benchmark (TruthfulQA) demonstrate our consistent effectiveness, e.g., up to 8.6 absolute points on TruthfulQA. We believe this study can improve our understanding of hallucinations and serve as a practical solution for hallucination mitigation.

New NLP benchmarks are urgently needed to align with the rapid development of large language models (LLMs). We present C-Eval, the first comprehensive Chinese evaluation suite designed to assess advanced knowledge and reasoning abilities of foundation models in a Chinese context. C-Eval comprises multiple-choice questions across four difficulty levels: middle school, high school, college, and professional. The questions span 52 diverse disciplines, ranging from humanities to science and engineering. C-Eval is accompanied by C-Eval Hard, a subset of very challenging subjects in C-Eval that requires advanced reasoning abilities to solve. We conduct a comprehensive evaluation of the most advanced LLMs on C-Eval, including both English- and Chinese-oriented models. Results indicate that only GPT-4 could achieve an average accuracy of over 60%, suggesting that there is still significant room for improvement for current LLMs. We anticipate C-Eval will help analyze important strengths and shortcomings of foundation models, and foster their development and growth for Chinese users.

As an efficient alternative to conventional full fine-tuning, parameter-efficient fine-tuning (PEFT) is becoming the prevailing method to adapt pretrained language models. In PEFT, a lightweight module is learned on each dataset while the underlying pretrained language model remains unchanged, resulting in multiple compact modules representing diverse skills when applied to various domains and tasks. In this paper, we propose to compose these parameter-efficient modules through linear arithmetic operations in the weight space, thereby integrating different module capabilities. Specifically, we first define an addition and negation operator for the module, and then further compose these two basic operators to perform flexible arithmetic. Our approach requires no additional training and enables highly flexible module composition. We apply different arithmetic operations to compose the parameter-efficient modules for (1) distribution generalization, (2) multi-tasking, (3) detoxifying, and (4) domain transfer. Additionally, we extend our approach to detoxify Alpaca-LoRA, the latest instruction-tuned large language model based on LLaMA. Empirical results demonstrate that our approach produces new and effective parameter-efficient modules that significantly outperform existing ones across all settings.