Large Language Models (LLMs) have demonstrated impressive mathematical reasoning capabilities, yet their performance remains brittle to minor variations in problem description and prompting strategy. Furthermore, reasoning is vulnerable to sampling-induced errors which autoregressive models must primarily address using self-correction via additionally-generated tokens. To better understand self-correction capabilities of recent models, we conduct experiments measuring models' ability to self-correct synthetic perturbations introduced into their Chain of Thought (CoT) reasoning. We observe robust single-utterance intrinsic self-correction behavior across a range of open-weight models and datasets, ranging from subtle, implicit corrections to explicit acknowledgments and corrections of errors. Our findings suggest that LLMs, including those not finetuned for long CoT, may possess stronger intrinsic self-correction capabilities than commonly shown in the literature. The presence of this ability suggests that recent "reasoning" model work involves amplification of traits already meaningfully present in models.

-- Vision-Language-Action (VLA) models have emerged as a prominent research area, showcasing significant potential across a variety of applications. However, their performance in endoscopy robotics, particularly endoscopy capsule robots that perform actions within the digestive system, remains unexplored. The integration of VLA models into endoscopy robots allows more intuitive and efficient interactions between human operators and medical devices, improving both diagnostic accuracy and treatment outcomes. By processing interleaved visual inputs, and textual instructions, CapsDT can infer corresponding robotic control signals to facilitate endoscopy tasks. In addition, we developed a capsule endoscopy robot system, a capsule robot controlled by a robotic arm-held magnet, addressing different levels of four endoscopy tasks and creating corresponding capsule robot datasets within the stomach simulator . Comprehensive evaluations on various robotic tasks indicate that CapsDT can serve as a robust vision-language generalist, achieving state-of-the-art performance in various levels of endoscopy tasks while achieving a 26.25% success rate in real-world simulation manipulation. I. INTRODUCTION Endoscopy, for both diagnostic and therapeutic interventions, provides direct visualization and treatment capabilities within the gastrointestinal (GI) tract [1], [2], [3].

Small language models (SLMs) struggle to learn complex reasoning behaviors, especially when high-quality traces are scarce or difficult to learn from. The standard training approach combines a supervised fine-tuning (SFT) stage, often to distill capabilities of a larger model, followed by a reinforcement learning (RL)stage such as Group Relative Policy Optimization (GRPO). In this paper, we investigate the fundamental limitations of this SFT + RL paradigm and propose methods to overcome them. Under a suitable theoretical model, we demonstrate that the SFT + RL strategy can fail completely when (1) the expert's traces are too difficult for the small model to express, or (2) the small model's initialization has exponentially small likelihood of success. To address these, we introduce BREAD: a GRPO variant that unifies the SFT and RL stages via partial expert guidance and branched rollouts. When self-generated traces fail, BREAD adaptively inserts short expert prefixes/hints, allowing the small model to complete the rest of the reasoning path, and ensuring that each update includes at least one successful trace. This mechanism both densifies the reward signal and induces a natural learning curriculum. BREAD requires fewer than 40% of ground-truth traces, consistently outperforming standard GRPO while speeding up the training by about 3 times. Importantly, we demonstrate that BREAD helps the model solve problems that are otherwise unsolvable by the SFT + RL strategy, highlighting how branched rollouts and expert guidance can substantially boost SLM reasoning.

The'findings' section of a radiology report is often detailed and lengthy, whereas the'impression' section is comparatively more compact and captures key diagnostic conclusions. This research explores the use of advanced abstractive summarization models to generate the concise'impression' from the'findings' section of a radiology report. We have used the publicly available MIMIC-CXR dataset. A comparative analysis is conducted on leading pre-trained and open-source large language models, including T5-base, BART-base, PEGASUS-x-base, ChatGPT-4, LLaMA-3-8B, and a custom Pointer Generator Network with a coverage mechanism. To ensure a thorough assessment, multiple evaluation metrics are employed, including ROUGE-1, ROUGE-2, ROUGE-L, METEOR, and BERTScore. By analyzing the performance of these models, this study identifies their respective strengths and limitations in the summarization of medical text. The findings of this paper provide helpful information for medical professionals who need automated summarization solutions in the healthcare sector.

Abstract--Multivariate time series forecasting (MTSF) is a critical task with broad applications in domains such as meteorology, transportation, and economics. Nevertheless, pervasive missing values caused by sensor failures or human errors significantly degrade forecasting accuracy . Prior efforts usually employ an impute-then-forecast paradigm, leading to suboptimal predictions due to error accumulation and misaligned objectives between the two stages. T o address this challenge, we propose the Collaborative Imputation-Forecasting Network (CoIFNet), a novel framework that unifies imputation and forecasting to achieve robust MTSF in the presence of missing values. Specifically, CoIFNet takes the observed values, mask matrix and timestamp embeddings as input, processing them sequentially through the Cross-Timestep Fusion (CTF) and Cross-Variate Fusion (CVF) modules to capture temporal dependencies that are robust to missing values. We provide theoretical justifications on how our CoIFNet learning objective improves the performance bound of MTSF with missing values. Through extensive experiments on challenging MSTF benchmarks, we demonstrate the effectiveness and computational efficiency of our proposed approach across diverse missing-data scenarios, e.g., CoIFNet outperforms the state-of-the-art method by 24.40 % ( 23.81 %) at a point (block) missing rate of 0.6, while improving memory and time efficiency by Our code is available at: https://github.com/KaiT By modeling complex temporal dependencies and inter-variate correlations, advanced MTSF methods [4], [5], [6] have demonstrated significant capabilities in predicting critical metrics and anticipating future trends.

Identifying inputs that trigger specific behaviours or latent features in language models could have a wide range of safety use cases. We investigate a class of methods capable of generating targeted, linguistically fluent inputs that activate specific latent features or elicit model behaviours. We formalise this approach as context modification and present ContextBench -- a benchmark with tasks assessing core method capabilities and potential safety applications. Our evaluation framework measures both elicitation strength (activation of latent features or behaviours) and linguistic fluency, highlighting how current state-of-the-art methods struggle to balance these objectives. We enhance Evolutionary Prompt Optimisation (EPO) with LLM-assistance and diffusion model inpainting, and demonstrate that these variants achieve state-of-the-art performance in balancing elicitation effectiveness and fluency.

We develop a framework to quantify the time-to-unsafe-sampling - the number of large language model (LLM) generations required to trigger an unsafe (e.g., toxic) response. Estimating this quantity is challenging, since unsafe responses are exceedingly rare in well-aligned LLMs, potentially occurring only once in thousands of generations. As a result, directly estimating time-to-unsafe-sampling would require collecting training data with a prohibitively large number of generations per prompt. However, with realistic sampling budgets, we often cannot generate enough responses to observe an unsafe outcome for every prompt, leaving the time-to-unsafe-sampling unobserved in many cases, making the estimation and evaluation tasks particularly challenging. To address this, we frame this estimation problem as one of survival analysis and develop a provably calibrated lower predictive bound (LPB) on the time-to-unsafe-sampling of a given prompt, leveraging recent advances in conformal prediction. Our key innovation is designing an adaptive, per-prompt sampling strategy, formulated as a convex optimization problem. The objective function guiding this optimized sampling allocation is designed to reduce the variance of the estimators used to construct the LPB, leading to improved statistical efficiency over naive methods that use a fixed sampling budget per prompt. Experiments on both synthetic and real data support our theoretical results and demonstrate the practical utility of our method for safety risk assessment in generative AI models.

Sampling from language models impacts the quality and diversity of outputs, affecting both research and real-world applications. Recently, Nguyen et al. 2024's "Turning Up the Heat: Min-p Sampling for Creative and Coherent LLM Outputs" introduced a new sampler called min-p, claiming it achieves superior quality and diversity over established samplers such as basic, top-k, and top-p sampling. The significance of these claims was underscored by the paper's recognition as the 18th highest-scoring submission to ICLR 2025 and selection for an Oral presentation. This paper conducts a comprehensive re-examination of the evidence supporting min-p and reaches different conclusions from the original paper's four lines of evidence. First, the original paper's human evaluations omitted data, conducted statistical tests incorrectly, and described qualitative feedback inaccurately; our reanalysis demonstrates min-p did not outperform baselines in quality, diversity, or a trade-off between quality and diversity; in response to our findings, the authors of the original paper conducted a new human evaluation using a different implementation, task, and rubric that nevertheless provides further evidence min-p does not improve over baselines. Second, comprehensively sweeping the original paper's NLP benchmarks reveals min-p does not surpass baselines when controlling for the number of hyperparameters. Third, the original paper's LLM-as-a-Judge evaluations lack methodological clarity and appear inconsistently reported. Fourth, community adoption claims (49k GitHub repositories, 1.1M GitHub stars) were found to be unsubstantiated, leading to their removal; the revised adoption claim remains misleading. We conclude that evidence presented in the original paper fails to support claims that min-p improves quality, diversity, or a trade-off between quality and diversity.

Historical geologic maps contain rich geospatial information, such as rock units, faults, folds, and bedding planes, that is critical for assessing mineral resources essential to renewable energy, electric vehicles, and national security. However, digitizing maps remains a labor-intensive and time-consuming task. We present DIGMAPPER, a modular, scalable system developed in collaboration with the United States Geological Survey (USGS) to automate the digitization of geologic maps. DIGMAPPER features a fully dockerized, workflow-orchestrated architecture that integrates state-of-the-art deep learning models for map layout analysis, feature extraction, and georeferencing. To overcome challenges such as limited training data and complex visual content, our system employs innovative techniques, including in-context learning with large language models, synthetic data generation, and transformer-based models. Evaluations on over 100 annotated maps from the DARPA-USGS dataset demonstrate high accuracy across polygon, line, and point feature extraction, and reliable georeferencing performance. Deployed at USGS, DIGMAPPER significantly accelerates the creation of analysis-ready geospatial datasets, supporting national-scale critical mineral assessments and broader geoscientific applications.

Due to an exponential increase in published research articles, it is impossible for individual scientists to read all publications, even within their own research field. In this work, we investigate the use of large language models (LLMs) for the purpose of extracting the main concepts and semantic information from scientific abstracts in the domain of materials science to find links that were not noticed by humans and thus to suggest inspiring near/mid-term future research directions. We show that LLMs can extract concepts more efficiently than automated keyword extraction methods to build a concept graph as an abstraction of the scientific literature. A machine learning model is trained to predict emerging combinations of concepts, i.e. new research ideas, based on historical data. We demonstrate that integrating semantic concept information leads to an increased prediction performance. The applicability of our model is demonstrated in qualitative interviews with domain experts based on individualized model suggestions. We show that the model can inspire materials scientists in their creative thinking process by predicting innovative combinations of topics that have not yet been investigated.