Accurate water level forecasting is crucial for managing ecosystems such as the Everglades, a subtropical wetland vital for flood mitigation, drought management, water resource planning, and biodiversity conservation. While recent advances in deep learning, particularly time series foundation models, have demonstrated success in general-domain forecasting, their application in hydrology remains underex-plored. Furthermore, they often struggle to generalize across diverse unseen datasets and domains, due to the lack of effective mechanisms for adaptation. To address this gap, we introduce Retrieval-Augmented Forecasting (RAF) into the hydrology domain, proposing a framework that retrieves historically analogous multivariate hydrological episodes to enrich the model input before forecasting. By maintaining an external archive of past observations, RAF identifies and incorporates relevant patterns from historical data, thereby enhancing contextual awareness and predictive accuracy without requiring the model for task-specific retraining or fine-tuning. Furthermore, we explore and compare both similarity-based and mutual information -based RAF methods. We conduct a comprehensive evaluation on real-world data from the Everglades, demonstrating that the RAF framework yields substantial improvements in water level forecasting accuracy. This study highlights the potential of RAF approaches in environmental hydrology and paves the way for broader adoption of adaptive AI methods by domain experts in ecosystem management.

Large language models (LLMs) already excel at writing code in high-resource languages such as Python and JavaScript, yet stumble on low-resource languages that remain essential to science and engineering. Besides the obvious shortage of pre-training data, post-training itself is a bottleneck: every new language seems to require new datasets, test harnesses, and reinforcement-learning (RL) infrastructure. We introduce Agnostics, a language-agnostic post-training pipeline that eliminates this per-language engineering. The key idea is to judge code solely by its externally observable behavior, so a single verifier can test solutions written in any language. Concretely, we (i) use an LLM to rewrite existing unit-test datasets into an I/O format, (ii) supply a short configuration that tells the verifier how to compile and run a target language, and (iii) apply reinforcement learning with verifiable rewards (RLVR) in a robust code execution environment. Applied to five low-resource languages--Lua, Julia, R, OCaml, and Fortran--Agnostics (1) improves Qwen-3 4B to performance that rivals other 16B-70B open-weight models; (2) scales cleanly to larger and diverse model families (Qwen-3 8B, DeepSeek Coder 6.7B Instruct, Phi 4 Mini); and (3) for ${\le} 16$B parameter models, sets new state-of-the-art pass@1 results on MultiPL-E and a new multi-language version LiveCodeBench that we introduce. We will release the language-agnostic training datasets (Ag-MBPP-X, Ag-Codeforces-X, Ag-LiveCodeBench-X), training code, and ready-to-use configurations, making RL post-training in any programming language as simple as editing a short YAML file.

Post-training quantization (PTQ) has become a crucial tool for reducing the memory and compute costs of modern deep neural networks, including large language models (LLMs). Among PTQ algorithms, the OPTQ framework-also known as GPTQ-has emerged as a leading method due to its computational efficiency and strong empirical performance. Despite its widespread adoption, however, OPTQ lacks rigorous quantitative theoretical guarantees. This paper presents the first quantitative error bounds for both deterministic and stochastic variants of OPTQ, as well as for Qronos, a recent related state-of-the-art PTQ algorithm. We analyze how OPTQ's iterative procedure induces quantization error and derive non-asymptotic 2-norm error bounds that depend explicitly on the calibration data and a regularization parameter that OPTQ uses. Our analysis provides theoretical justification for several practical design choices, including the widely used heuristic of ordering features by decreasing norm, as well as guidance for selecting the regularization parameter. For the stochastic variant, we establish stronger infinity-norm error bounds, which enable control over the required quantization alphabet and are particularly useful for downstream layers and nonlinearities. Finally, we extend our analysis to Qronos, providing new theoretical bounds, for both its deterministic and stochastic variants, that help explain its empirical advantages.

Reinforcement learning (RL) has become a key technique for enhancing the reasoning abilities of large language models (LLMs), with policy-gradient algorithms dominating the post-training stage because of their efficiency and effectiveness. However, most existing benchmarks evaluate large-language-model reasoning under idealized settings, overlooking performance in realistic, non-ideal scenarios. We identify three representative non-ideal scenarios with practical relevance: summary inference, fine-grained noise suppression, and contextual filtering. We introduce a new research direction guided by brain-science findings that human reasoning remains reliable under imperfect inputs. We formally define and evaluate these challenging scenarios. We fine-tune three LLMs and a state-of-the-art large vision-language model (LVLM) using RL with a representative policy-gradient algorithm and then test their performance on eight public datasets. Our results reveal that while RL fine-tuning improves baseline reasoning under idealized settings, performance declines significantly across all three non-ideal scenarios, exposing critical limitations in advanced reasoning capabilities. Although we propose a scenario-specific remediation method, our results suggest current methods leave these reasoning deficits largely unresolved. This work highlights that the reasoning abilities of large models are often overstated and underscores the importance of evaluating models under non-ideal scenarios. The code and data will be released at XXXX.

Autonomous web-based geographical information systems (AWebGIS) aim to perform geospatial operations from natural language input, providing intuitive, intelligent, and hands-free interaction. However, most current solutions rely on cloud-based large language models (LLMs), which require continuous internet access and raise users' privacy and scalability issues due to centralized server processing. This study compares three approaches to enabling AWebGIS: (1) a fully-automated online method using cloud-based LLMs (e.g., Cohere); (2) a semi-automated offline method using classical machine learning classifiers such as support vector machine and random forest; and (3) a fully autonomous offline (client-side) method based on a fine-tuned small language model (SLM), specifically T5-small model, executed in the client's web browser. The third approach, which leverages SLMs, achieved the highest accuracy among all methods, with an exact matching accuracy of 0.93, Levenshtein similarity of 0.99, and recall-oriented understudy for gisting evaluation ROUGE-1 and ROUGE-L scores of 0.98. Crucially, this client-side computation strategy reduces the load on backend servers by offloading processing to the user's device, eliminating the need for server-based inference. These results highlight the feasibility of browser-executable models for AWebGIS solutions.

Logging is essential in software development, helping developers monitor system behavior and aiding in debugging applications. Given the ability of large language models (LLMs) to generate natural language and code, researchers are exploring their potential to generate log statements. However, prior work focuses on evaluating logs introduced in code functions, leaving file-level log generation underexplored -- especially in machine learning (ML) applications, where comprehensive logging can enhance reliability. In this study, we evaluate the capacity of GPT-4o mini as a case study to generate log statements for ML projects at file level. We gathered a set of 171 ML repositories containing 4,073 Python files with at least one log statement. We identified and removed the original logs from the files, prompted the LLM to generate logs for them, and evaluated both the position of the logs and log level, variables, and text quality of the generated logs compared to human-written logs. In addition, we manually analyzed a representative sample of generated logs to identify common patterns and challenges. We find that the LLM introduces logs in the same place as humans in 63.91% of cases, but at the cost of a high overlogging rate of 82.66%. Furthermore, our manual analysis reveals challenges for file-level logging, which shows overlogging at the beginning or end of a function, difficulty logging within large code blocks, and misalignment with project-specific logging conventions. While the LLM shows promise for generating logs for complete files, these limitations remain to be addressed for practical implementation.

This study examined whether embedding LLM-guided reflection prompts in an interactive AI-generated podcast improved learning and user experience compared to a version without prompts. Thirty-six undergraduates participated, and while learning outcomes were similar across conditions, reflection prompts reduced perceived attractiveness, highlighting a call for more research on reflective interactivity design.

Large pre-trained DNA language models such as DNABERT-2, Nucleotide Transformer, and HyenaDNA have demonstrated strong performance on various genomic benchmarks. However, most applications rely on expensive fine-tuning, which works best when the training and test data share a similar distribution. In this work, we investigate whether task-specific fine-tuning is always necessary. We show that simple embedding-based pipelines that extract fixed representations from these models and feed them into lightweight classifiers can achieve competitive performance. In evaluation settings with different data distributions, embedding-based methods often outperform fine-tuning while reducing inference time by 10x to 20x. Our results suggest that embedding extraction is not only a strong baseline but also a more generalizable and efficient alternative to fine-tuning, especially for deployment in diverse or unseen genomic contexts. For example, in enhancer classification, HyenaDNA embeddings combined with zCurve achieve 0.68 accuracy (vs. 0.58 for fine-tuning), with an 88% reduction in inference time and over 8x lower carbon emissions (0.02 kg vs. 0.17 kg CO2). In non-TATA promoter classification, DNABERT-2 embeddings with zCurve or GC content reach 0.85 accuracy (vs. 0.89 with fine-tuning) with a 22x lower carbon footprint (0.02 kg vs. 0.44 kg CO2). These results show that embedding-based pipelines offer over 10x better carbon efficiency while maintaining strong predictive performance. The code is available here: https://github.com/NIRJHOR-DATTA/EMBEDDING-IS-ALMOST-ALL-YOU-NEED.

Reinforcement learning (RL)-based dynamic treatment regimes (DTRs) hold promise for automating complex clinical decision-making, yet their practical deployment remains hindered by the intensive engineering required to inject clinical knowledge and ensure patient safety. Recent advancements in large language models (LLMs) suggest a complementary approach, where implicit prior knowledge and clinical heuristics are naturally embedded through linguistic prompts without requiring environment-specific training. In this study, we rigorously evaluate open-source LLMs as dynamic insulin dosing agents in an in silico Type 1 diabetes simulator, comparing their zero-shot inference performance against small neural network-based RL agents (SRAs) explicitly trained for the task. Our results indicate that carefully designed zero-shot prompts enable smaller LLMs (e.g., Qwen2.5-7B) to achieve comparable or superior clinical performance relative to extensively trained SRAs, particularly in stable patient cohorts. However, LLMs exhibit notable limitations, such as overly aggressive insulin dosing when prompted with chain-of-thought (CoT) reasoning, highlighting critical failure modes including arithmetic hallucination, temporal misinterpretation, and inconsistent clinical logic. Incorporating explicit reasoning about latent clinical states (e.g., meals) yielded minimal performance gains, underscoring the current model's limitations in capturing complex, hidden physiological dynamics solely through textual inference. Our findings advocate for cautious yet optimistic integration of LLMs into clinical workflows, emphasising the necessity of targeted prompt engineering, careful validation, and potentially hybrid approaches that combine linguistic reasoning with structured physiological modelling to achieve safe, robust, and clinically effective decision-support systems.

Cell type annotation is a fundamental step in the analysis of single-cell RNA sequencing (scRNA-seq) data. In practice, human experts often rely on the structure revealed by principal component analysis (PCA) followed by $k$-nearest neighbor ($k$-NN) graph construction to guide annotation. While effective, this process is labor-intensive and does not scale to large datasets. Recent advances in CLIP-style models offer a promising path toward automating cell type annotation. By aligning scRNA-seq profiles with natural language descriptions, models like LangCell enable zero-shot annotation. While LangCell demonstrates decent zero-shot performance, its predictions remain suboptimal, particularly in achieving consistent accuracy across all cell types. In this paper, we propose to refine the zero-shot logits produced by LangCell through a graph-regularized optimization framework. By enforcing local consistency over the task-specific PCA-based k-NN graph, our method combines the scalability of the pre-trained models with the structural robustness relied upon in expert annotation. We evaluate our approach on 14 annotated human scRNA-seq datasets from 4 distinct studies, spanning 11 organs and over 200,000 single cells. Our method consistently improves zero-shot annotation accuracy, achieving accuracy gains of up to 10%. Further analysis showcase the mechanism by which GRIT effectively propagates correct signals through the graph, pulling back mislabeled cells toward more accurate predictions. The method is training-free, model-agnostic, and serves as a simple yet effective plug-in for enhancing automated cell type annotation in practice.