With the growing demand for personalized AI solutions, customized LLMs have become a preferred choice for businesses and individuals, driving the deployment of millions of AI agents across various platforms, e.g., GPT Store hosts over 3 million customized GPTs. Their popularity is partly driven by advanced reasoning capabilities, such as Chain-of-Thought, which enhance their ability to tackle complex tasks. However, their rapid proliferation introduces new vulnerabilities, particularly in reasoning processes that remain largely unexplored. We introduce DarkMind, a novel backdoor attack that exploits the reasoning capabilities of customized LLMs. Designed to remain latent, DarkMind activates within the reasoning chain to covertly alter the final outcome. Unlike existing attacks, it operates without injecting triggers into user queries, making it a more potent threat. We evaluate DarkMind across eight datasets covering arithmetic, commonsense, and symbolic reasoning domains, using five state-of-the-art LLMs with five distinct trigger implementations. Our results demonstrate DarkMind effectiveness across all scenarios, underscoring its impact. Finally, we explore potential defense mechanisms to mitigate its risks, emphasizing the need for stronger security measures.

Large language models (LLMs) have excelled in various applications, yet serving them at scale is challenging due to their substantial resource demands and high latency. Our real-world studies reveal that over 60% of user requests to LLMs have semantically similar counterparts, suggesting the potential for knowledge sharing among requests. However, naively caching and reusing past responses leads to large quality degradation. In this paper, we introduce EchoLM, an in-context caching system that leverages historical requests as examples to guide response generation, enabling selective offloading of requests to more efficient LLMs. However, enabling this real-time knowledge transfer leads to intricate tradeoffs between response quality, latency, and system throughput at scale. For a new request, EchoLM identifies similar, high-utility examples and efficiently prepends them to the input for better response. At scale, EchoLM adaptively routes requests to LLMs of varying capabilities, accounting for response quality and serving loads. EchoLM employs a cost-aware cache replay mechanism to improve example quality and coverage offline, maximizing cache utility and runtime efficiency. Evaluations on millions of open-source requests demonstrate that EchoLM has a throughput improvement of 1.4-5.9x while reducing latency by 28-71% without hurting response quality on average.

Foundation models are widely employed in medical image analysis, due to their high adaptability and generalizability for downstream tasks. With the increasing number of foundation models being released, model selection has become an important issue. In this work, we study the capabilities of foundation models in medical image classification tasks by conducting a benchmark study on the MedMNIST dataset. Specifically, we adopt various foundation models ranging from convolutional to Transformer-based models and implement both end-to-end training and linear probing for all classification tasks. The results demonstrate the significant potential of these pre-trained models when transferred for medical image classification. We further conduct experiments with different image sizes and various sizes of training data. By analyzing all the results, we provide preliminary, yet useful insights and conclusions on this topic.

One of the most widely used methods to evaluate LLMs are Multiple Choice Question (MCQ) tests. MCQ benchmarks enable the testing of LLM knowledge on almost any topic at scale as the results can be processed automatically. To help the LLM answer, a few examples called few shots can be included in the prompt. Moreover, the LLM can be asked to answer the question directly with the selected option or to first provide the reasoning and then the selected answer, which is known as chain of thought. In addition to checking whether the selected answer is correct, the evaluation can look at the LLM-estimated probability of its response as an indication of the confidence of the LLM in the response. In this paper, we study how the LLM confidence in its answer depends on whether the model has been asked to answer directly or to provide the reasoning before answering. The results of the evaluation of questions on a wide range of topics in seven different models show that LLMs are more confident in their answers when they provide reasoning before the answer. This occurs regardless of whether the selected answer is correct. Our hypothesis is that this behavior is due to the reasoning that modifies the probability of the selected answer, as the LLM predicts the answer based on the input question and the reasoning that supports the selection made. Therefore, LLM estimated probabilities seem to have intrinsic limitations that should be understood in order to use them in evaluation procedures. Interestingly, the same behavior has been observed in humans, for whom explaining an answer increases confidence in its correctness.

In knowledge-intensive tasks, especially in high-stakes domains like medicine and law, it is critical not only to retrieve relevant information but also to provide causal reasoning and explainability. Large language models (LLMs) have achieved remarkable performance in natural language understanding and generation tasks. However, they often suffer from limitations such as difficulty in incorporating new knowledge, generating hallucinations, and explaining their reasoning process. To address these challenges, integrating knowledge graphs with Graph Retrieval-Augmented Generation (Graph RAG) has emerged as an effective solution. Traditional Graph RAG methods often rely on simple graph traversal or semantic similarity, which do not capture causal relationships or align well with the model's internal reasoning steps. This paper proposes a novel pipeline that filters large knowledge graphs to emphasize cause-effect edges, aligns the retrieval process with the model's chain-of-thought (CoT), and enhances reasoning through multi-stage path improvements. Experiments on medical question-answering tasks show consistent gains, with up to a 10\% absolute improvement across multiple large language models (LLMs). This approach demonstrates the value of combining causal reasoning with stepwise retrieval, leading to more interpretable and logically grounded solutions for complex queries.

Large language models (LLMs) often exhibit gender bias, posing challenges for their safe deployment. Existing methods to mitigate bias lack a comprehensive understanding of its mechanisms or compromise the model's core capabilities. To address these issues, we propose the CommonWords dataset, to systematically evaluate gender bias in LLMs. Our analysis reveals pervasive bias across models and identifies specific neuron circuits, including gender neurons and general neurons, responsible for this behavior. Notably, editing even a small number of general neurons can disrupt the model's overall capabilities due to hierarchical neuron interactions. Based on these insights, we propose an interpretable neuron editing method that combines logit-based and causal-based strategies to selectively target biased neurons. Experiments on five LLMs demonstrate that our method effectively reduces gender bias while preserving the model's original capabilities, outperforming existing fine-tuning and editing approaches. Our findings contribute a novel dataset, a detailed analysis of bias mechanisms, and a practical solution for mitigating gender bias in LLMs.

The rampant spread of cyberbullying content poses a growing threat to societal well-being. However, research on cyberbullying detection in Chinese remains underdeveloped, primarily due to the lack of comprehensive and reliable datasets. Notably, no existing Chinese dataset is specifically tailored for cyberbullying detection. Moreover, while comments play a crucial role within sessions, current session-based datasets often lack detailed, fine-grained annotations at the comment level. To address these limitations, we present a novel Chinese cyber-bullying dataset, termed SCCD, which consists of 677 session-level samples sourced from a major social media platform Weibo. Moreover, each comment within the sessions is annotated with fine-grained labels rather than conventional binary class labels. Empirically, we evaluate the performance of various baseline methods on SCCD, highlighting the challenges for effective Chinese cyberbullying detection.

Automated grading has become an essential tool in education technology due to its ability to efficiently assess large volumes of student work, provide consistent and unbiased evaluations, and deliver immediate feedback to enhance learning. However, current systems face significant limitations, including the need for large datasets in few-shot learning methods, a lack of personalized and actionable feedback, and an overemphasis on benchmark performance rather than student experience. To address these challenges, we propose a Zero-Shot Large Language Model (LLM)-Based Automated Assignment Grading (AAG) system. This framework leverages prompt engineering to evaluate both computational and explanatory student responses without requiring additional training or fine-tuning. The AAG system delivers tailored feedback that highlights individual strengths and areas for improvement, thereby enhancing student learning outcomes. Our study demonstrates the system's effectiveness through comprehensive evaluations, including survey responses from higher education students that indicate significant improvements in motivation, understanding, and preparedness compared to traditional grading methods. The results validate the AAG system's potential to transform educational assessment by prioritizing learning experiences and providing scalable, high-quality feedback.

As vector representations have been pivotal in advancing natural language processing (NLP), some prior research has concentrated on creating embedding techniques for mathematical expressions by leveraging mathematically equivalent expressions. While effective, these methods are limited by the training data. In this work, we propose augmenting prior algorithms with larger synthetic dataset, using a novel e-graph-based generation scheme. This new mathematical dataset generation scheme, E-Gen, improves upon prior dataset-generation schemes that are limited in size and operator types. We use this dataset to compare embedding models trained with two methods: (1) training the model to generate mathematically equivalent expressions, and (2) training the model using contrastive learning to group mathematically equivalent expressions explicitly. We evaluate the embeddings generated by these methods against prior work on both in-distribution and out-of-distribution language processing tasks. Finally, we compare the performance of our embedding scheme against state-of-the-art large language models and demonstrate that embedding-based language processing methods perform better than LLMs on several tasks, demonstrating the necessity of optimizing embedding methods for the mathematical data modality.

Research question answering requires accurate retrieval and contextual understanding of scientific literature. However, current Retrieval-Augmented Generation (RAG) methods often struggle to balance complex document relationships with precise information retrieval. In this paper, we introduce Contextualized Graph Retrieval-Augmented Generation (CG-RAG), a novel framework that integrates sparse and dense retrieval signals within graph structures to enhance retrieval efficiency and subsequently improve generation quality for research question answering. First, we propose a contextual graph representation for citation graphs, effectively capturing both explicit and implicit connections within and across documents. Next, we introduce Lexical-Semantic Graph Retrieval (LeSeGR), which seamlessly integrates sparse and dense retrieval signals with graph encoding. It bridges the gap between lexical precision and semantic understanding in citation graph retrieval, demonstrating generalizability to existing graph retrieval and hybrid retrieval methods. Finally, we present a context-aware generation strategy that utilizes the retrieved graph-structured information to generate precise and contextually enriched responses using large language models (LLMs). Extensive experiments on research question answering benchmarks across multiple domains demonstrate that our CG-RAG framework significantly outperforms RAG methods combined with various state-of-the-art retrieval approaches, delivering superior retrieval accuracy and generation quality.