Thinking Large Language Models (LLMs) generate explicit intermediate reasoning traces before final answers, potentially improving transparency, interpretability, and solution accuracy for code generation. However, the quality of these reasoning chains remains underexplored. We present a comprehensive empirical study examining the reasoning process and quality of thinking LLMs for code generation. We evaluate six state-of-the-art reasoning LLMs (DeepSeek-R1, OpenAI-o3-mini, Claude-3.7-Sonnet-Thinking, Gemini-2.0-Flash-Thinking, Gemini-2.5-Flash, and Qwen-QwQ) across 100 code generation tasks of varying difficulty from BigCodeBench. We quantify reasoning-chain structure through step counts and verbosity, conduct controlled step-budget adjustments, and perform a 21-participant human evaluation across three dimensions: efficiency, logical correctness, and completeness. Our step-count interventions reveal that targeted step increases can improve resolution rates for certain models/tasks, while modest reductions often preserve success on standard tasks, rarely on hard ones. Through systematic analysis, we develop a reasoning-problematic taxonomy, identifying completeness as the dominant failure mode. Task complexity significantly impacts reasoning quality; hard problems are substantially more prone to incompleteness than standard tasks. Our stability analysis demonstrates that thinking LLMs maintain consistent logical structures across computational effort levels and can self-correct previous errors. This study provides new insights into the strengths and limitations of current thinking LLMs in software engineering.

Tabular Prior-Data Fitted Network (TabPFN) is a foundation model designed for small to medium-sized tabular data, which has attracted much attention recently. This paper investigates the application of TabPFN in Combinatorial Optimization (CO) problems. The aim is to lessen challenges in time and data-intensive training requirements often observed in using traditional methods including exact and heuristic algorithms, Machine Learning (ML)-based models, to solve CO problems. Proposing possibly the first ever application of TabPFN for such a purpose, we adapt and fine-tune the TabPFN model to solve the Travelling Salesman Problem (TSP), one of the most well-known CO problems. Specifically, we adopt the node-based approach and the node-predicting adaptation strategy to construct the entire TSP route. Our evaluation with varying instance sizes confirms that TabPFN requires minimal training, adapts to TSP using a single sample, performs better generalization across varying TSP instance sizes, and reduces performance degradation. Furthermore, the training process with adaptation and fine-tuning is completed within minutes. The methodology leads to strong solution quality even without post-processing and achieves performance comparable to other models with post-processing refinement. Our findings suggest that the TabPFN model is a promising approach to solve structured and CO problems efficiently under training resource constraints and rapid deployment requirements.

Hallucination in large language models (LLMs) remains a critical barrier to their safe deployment. Existing tool-augmented hallucination detection methods require pre-defined fixed verification strategies, which are crucial to the quality and effectiveness of tool calls. Some methods directly employ powerful closed-source LLMs such as GPT-4 as detectors, which are effective but too costly. To mitigate the cost issue, some methods adopt the teacher-student architecture and finetune open-source small models as detectors via agent tuning. However, these methods are limited by fixed strategies. When faced with a dynamically changing execution environment, they may lack adaptability and inappropriately call tools, ultimately leading to detection failure. To address the problem of insufficient strategy adaptability, we propose the innovative ``Learning to Evaluate and Adaptively Plan''(LEAP) framework, which endows an efficient student model with the dynamic learning and proactive correction capabilities of the teacher model. Specifically, our method formulates the hallucination detection problem as a dynamic strategy learning problem. We first employ a teacher model to generate trajectories within the dynamic learning loop and dynamically adjust the strategy based on execution failures. We then distill this dynamic planning capability into an efficient student model via agent tuning. Finally, during strategy execution, the student model adopts a proactive correction mechanism, enabling it to propose, review, and optimize its own verification strategies before execution. We demonstrate through experiments on three challenging benchmarks that our LEAP-tuned model outperforms existing state-of-the-art methods.

Abstract--The ability of large language models (LLMs) to recall and retrieve information from long contexts is critical for many real-world applications. Prior work (Liu et al., 2023) reported that LLMs suffer significant drops in retrieval accuracy for facts placed in the middle of large contexts, an effect known as "Lost in the Middle" (LITM). We find the model Gemini 2.5 Flash can answer needle-in-a-haystack questions with great accuracy regardless of document position including when the document is nearly at the input context limit. Our results suggest that the "Lost in the Middle" effect is not present for simple factoid Q&A in Gemini 2.5 Flash, indicating substantial improvements in long-context retrieval. Large language models (LLMs) have rapidly advanced in their ability to process and reason over long textual contexts, enabling applications in summarization, retrieval-augmented Q&A, document understanding, and more.

Symbolic regression seeks to uncover physical laws from experimental data by searching for closed-form expressions, which is an important task in AI-driven scientific discovery. Yet the exponential growth of the search space of expression renders the task computationally challenging. A promising yet underexplored direction for reducing the effective search space and accelerating training lies in symbolic equivalence: many expressions, although syntactically different, define the same function -- for example, $\log(x_1^2x_2^3)$, $\log(x_1^2)+\log(x_2^3)$, and $2\log(x_1)+3\log(x_2)$. Existing algorithms treat such variants as distinct outputs, leading to redundant exploration and slow learning. We introduce EGG-SR, a unified framework that integrates equality graphs (e-graphs) into diverse symbolic regression algorithms, including Monte Carlo Tree Search (MCTS), deep reinforcement learning (DRL), and large language models (LLMs). EGG-SR compactly represents equivalent expressions through the proposed EGG module, enabling more efficient learning by: (1) pruning redundant subtree exploration in EGG-MCTS, (2) aggregating rewards across equivalence classes in EGG-DRL, and (3) enriching feedback prompts in EGG-LLM. Under mild assumptions, we show that embedding e-graphs tightens the regret bound of MCTS and reduces the variance of the DRL gradient estimator. Empirically, EGG-SR consistently enhances multiple baselines across challenging benchmarks, discovering equations with lower normalized mean squared error than state-of-the-art methods. Code implementation is available at: https://www.github.com/jiangnanhugo/egg-sr.

Alzheimer's disease is a prevalent neurodegenerative disorder for which early detection is critical. Handwriting-often disrupted in prodromal AD-provides a non-invasive and cost-effective window into subtle motor and cognitive decline. Existing handwriting-based AD studies, mostly relying on online trajectories and hand-crafted features, have not systematically examined how task type influences diagnostic performance and cross-task generalization. Meanwhile, large-scale vision language models have demonstrated remarkable zero or few-shot anomaly detection in natural images and strong adaptability across medical modalities such as chest X-ray and brain MRI. However, handwriting-based disease detection remains largely unexplored within this paradigm. To close this gap, we introduce a lightweight Cross-Layer Fusion Adapter framework that repurposes CLIP for handwriting-based AD screening. CLFA implants multi-level fusion adapters within the visual encoder to progressively align representations toward handwriting-specific medical cues, enabling prompt-free and efficient zero-shot inference. Using this framework, we systematically investigate cross-task generalization-training on a specific handwriting task and evaluating on unseen ones-to reveal which task types and writing patterns most effectively discriminate AD. Extensive analyses further highlight characteristic stroke patterns and task-level factors that contribute to early AD identification, offering both diagnostic insights and a benchmark for handwriting-based cognitive assessment.

With the development of cloud computing, the number of Web APIs has increased dramatically, further intensifying the demand for efficient Web API recommendation. Despite the demonstrated success of previous Web API recommendation solutions, two critical challenges persist: 1) a fixed top-N recommendation that cannot accommodate the varying API cardinality requirements of different mashups, and 2) these methods output only ranked API lists without accompanying reasons, depriving users of understanding the recommendation. To address these challenges, we propose WAR-Re, an LLM-based model for Web API recommendation with semantic reasoning for justification. WAR-Re leverages special start and stop tokens to handle the first challenge and uses two-stage training: supervised fine-tuning and reinforcement learning via Group Relative Policy Optimization (GRPO) to enhance the model's ability in both tasks. Comprehensive experimental evaluations on the ProgrammableWeb dataset demonstrate that WAR-Re achieves a gain of up to 21.59\% over the state-of-the-art baseline model in recommendation accuracy, while consistently producing high-quality semantic reasons for recommendations.

In today's landscape, Mixture of Experts (MoE) is a crucial architecture that has been used by many of the most advanced models. One of the major challenges of MoE models is that they usually require much more memory than their dense counterparts due to their unique architecture, and hence are harder to deploy in environments with limited GPU memory, such as edge devices. MoE offloading is a promising technique proposed to overcome this challenge, especially if it is enhanced with caching and pre-fetching, but prior work stopped at suboptimal caching algorithm and offered limited insights. In this work, we study MoE offloading in depth and make the following contributions: 1. We analyze the expert activation and LRU caching behavior in detail and provide traces. 2. We propose LFU caching optimization based on our analysis and obtain strong improvements from LRU. 3. We implement and experiment speculative expert pre-fetching, providing detailed trace showing its huge potential . 4. In addition, our study extensively covers the behavior of the MoE architecture itself, offering information on the characteristic of the gating network and experts. This can inspire future work on the interpretation of MoE models and the development of pruning techniques for MoE architecture with minimal performance loss.

Prompt injection attacks pose a critical threat to large language models (LLMs), with prior work focusing on cutting-edge LLM applications like personal copilots. In contrast, simpler LLM applications, such as customer service chatbots, are widespread on the web, yet their security posture and exposure to such attacks remain poorly understood. These applications often rely on third-party chatbot plugins that act as intermediaries to commercial LLM APIs, offering non-expert website builders intuitive ways to customize chatbot behaviors. To bridge this gap, we present the first large-scale study of 17 third-party chatbot plugins used by over 10,000 public websites, uncovering previously unknown prompt injection risks in practice. First, 8 of these plugins (used by 8,000 websites) fail to enforce the integrity of the conversation history transmitted in network requests between the website visitor and the chatbot. This oversight amplifies the impact of direct prompt injection attacks by allowing adversaries to forge conversation histories (including fake system messages), boosting their ability to elicit unintended behavior (e.g., code generation) by 3 to 8x. Second, 15 plugins offer tools, such as web-scraping, to enrich the chatbot's context with website-specific content. However, these tools do not distinguish the website's trusted content (e.g., product descriptions) from untrusted, third-party content (e.g., customer reviews), introducing a risk of indirect prompt injection. Notably, we found that ~13% of e-commerce websites have already exposed their chatbots to third-party content. We systematically evaluate both vulnerabilities through controlled experiments grounded in real-world observations, focusing on factors such as system prompt design and the underlying LLM. Our findings show that many plugins adopt insecure practices that undermine the built-in LLM safeguards.

Robotic grasping is a fundamental capability for autonomous manipulation; however, most existing methods rely on large-scale expert annotations and necessitate retraining to handle new objects. We present VLAD-Grasp, a Vision-Language model Assisted zero-shot approach for Detecting grasps. From a single RGB-D image, our method (1) prompts a large vision-language model to generate a goal image where a straight rod "impales" the object, representing an antipodal grasp, (2) predicts depth and segmentation to lift this generated image into 3D, and (3) aligns generated and observed object point clouds via principal component analysis and correspondence-free optimization to recover an executable grasp pose. Unlike prior work, our approach is training-free and does not rely on curated grasp datasets. Despite this, VLAD-Grasp achieves performance that is competitive with or superior to that of state-of-the-art supervised models on the Cornell and Jacquard datasets. We further demonstrate zero-shot generalization to novel real-world objects on a Franka Research 3 robot, highlighting vision-language foundation models as powerful priors for robotic manipulation.