In this paper we look at the ability of recent large language models (LLMs) at solving mathematical problems in combinatorics. We compare models LLaMA-2, LLaMA-3.1, GPT-4, and Mixtral against each other and against human pupils and undergraduates with prior experience in mathematical olympiads. To facilitate these comparisons we introduce the Combi-Puzzles dataset, which contains 125 problem variants based on 25 combinatorial reasoning problems. Each problem is presented in one of five distinct forms, created by systematically manipulating the problem statements through adversarial additions, numeric parameter changes, and linguistic obfuscation. Our variations preserve the mathematical core and are designed to measure the generalisability of LLM problem-solving abilities, while also increasing confidence that problems are submitted to LLMs in forms that have not been seen as training instances. We found that a model based on GPT-4 outperformed all other models in producing correct responses, and performed significantly better in the mathematical variation of the problems than humans. We also found that modifications to problem statements significantly impact the LLM's performance, while human performance remains unaffected.

Training medical personnel using standardized patients (SPs) remains a complex challenge, requiring extensive domain expertise and role-specific practice. Most research on Large Language Model (LLM)-based simulated patients focuses on improving data retrieval accuracy or adjusting prompts through human feedback. However, this focus has overlooked the critical need for patient agents to learn a standardized presentation pattern that transforms data into human-like patient responses through unsupervised simulations. To address this gap, we propose EvoPatient, a novel simulated patient framework in which a patient agent and doctor agents simulate the diagnostic process through multi-turn dialogues, simultaneously gathering experience to improve the quality of both questions and answers, ultimately enabling human doctor training. Extensive experiments on various cases demonstrate that, by providing only overall SP requirements, our framework improves over existing reasoning methods by more than 10% in requirement alignment and better human preference, while achieving an optimal balance of resource consumption after evolving over 200 cases for 10 hours, with excellent generalizability. The code will be available at https://github.com/ZJUMAI/EvoPatient.

This paper presents the systems we developed for RIRAG-2025, a shared task that requires answering regulatory questions by retrieving relevant passages. The generated answers are evaluated using RePASs, a reference-free and model-based metric. Our systems use a combination of three retrieval models and a reranker. We show that by exploiting a neural component of RePASs that extracts important sentences ('obligations') from the retrieved passages, we achieve a dubiously high score (0.947), even though the answers are directly extracted from the retrieved passages and are not actually generated answers. We then show that by selecting the answer with the best RePASs among a few generated alternatives and then iteratively refining this answer by reducing contradictions and covering more obligations, we can generate readable, coherent answers that achieve a more plausible and relatively high score (0.639).

Objectives: Compare qualitative coding of instruction tuned large language models (IT-LLMs) against human coders in classifying the presence or absence of vulnerability in routinely collected unstructured text that describes police-public interactions. Evaluate potential bias in IT-LLM codings. Methods: Analyzing publicly available text narratives of police-public interactions recorded by Boston Police Department, we provide humans and IT-LLMs with qualitative labelling codebooks and compare labels generated by both, seeking to identify situations associated with (i) mental ill health; (ii) substance misuse; (iii) alcohol dependence; and (iv) homelessness. We explore multiple prompting strategies and model sizes, and the variability of labels generated by repeated prompts. Additionally, to explore model bias, we utilize counterfactual methods to assess the impact of two protected characteristics - race and gender - on IT-LLM classification. Results: Results demonstrate that IT-LLMs can effectively support human qualitative coding of police incident narratives. While there is some disagreement between LLM and human generated labels, IT-LLMs are highly effective at screening narratives where no vulnerabilities are present, potentially vastly reducing the requirement for human coding. Counterfactual analyses demonstrate that manipulations to both gender and race of individuals described in narratives have very limited effects on IT-LLM classifications beyond those expected by chance. Conclusions: IT-LLMs offer effective means to augment human qualitative coding in a way that requires much lower levels of resource to analyze large unstructured datasets. Moreover, they encourage specificity in qualitative coding, promote transparency, and provide the opportunity for more standardized, replicable approaches to analyzing large free-text police data sources.

In-context learning (ICL) enables Large Language Models (LLMs) to perform tasks using few demonstrations, facilitating task adaptation when labeled examples are hard to obtain. However, ICL is sensitive to the choice of demonstrations, and it remains unclear which demonstration attributes enable in-context generalization. In this work, we conduct a perturbation study of in-context demonstrations for low-resource Named Entity Detection (NED). Our surprising finding is that in-context demonstrations with partially correct annotated entity mentions can be as effective for task transfer as fully correct demonstrations. Based off our findings, we propose Pseudo-annotated In-Context Learning (PICLe), a framework for in-context learning with noisy, pseudo-annotated demonstrations. PICLe leverages LLMs to annotate many demonstrations in a zero-shot first pass. We then cluster these synthetic demonstrations, sample specific sets of in-context demonstrations from each cluster, and predict entity mentions using each set independently. Finally, we use self-verification to select the final set of entity mentions. We evaluate PICLe on five biomedical NED datasets and show that, with zero human annotation, PICLe outperforms ICL in low-resource settings where limited gold examples can be used as in-context demonstrations.

Diffusion Transformers have emerged as the preeminent models for a wide array of generative tasks, demonstrating superior performance and efficacy across various applications. The promising results come at the cost of slow inference, as each denoising step requires running the whole transformer model with a large amount of parameters. In this paper, we show that performing the full computation of the model at each diffusion step is unnecessary, as some computations can be skipped by lazily reusing the results of previous steps. Furthermore, we show that the lower bound of similarity between outputs at consecutive steps is notably high, and this similarity can be linearly approximated using the inputs. To verify our demonstrations, we propose the \textbf{LazyDiT}, a lazy learning framework that efficiently leverages cached results from earlier steps to skip redundant computations. Specifically, we incorporate lazy learning layers into the model, effectively trained to maximize laziness, enabling dynamic skipping of redundant computations. Experimental results show that LazyDiT outperforms the DDIM sampler across multiple diffusion transformer models at various resolutions. Furthermore, we implement our method on mobile devices, achieving better performance than DDIM with similar latency.

Code completion has become an essential tool for daily software development. Existing evaluation benchmarks often employ static methods that do not fully capture the dynamic nature of real-world coding environments and face significant challenges, including limited context length, reliance on superficial evaluation metrics, and potential overfitting to training datasets. In this work, we introduce a novel framework for enhancing code completion in software development through the creation of a repository-level benchmark ExecRepoBench and the instruction corpora Repo-Instruct, aim at improving the functionality of open-source large language models (LLMs) in real-world coding scenarios that involve complex interdependencies across multiple files. ExecRepoBench includes 1.2K samples from active Python repositories. Plus, we present a multi-level grammar-based completion methodology conditioned on the abstract syntax tree to mask code fragments at various logical units (e.g. statements, expressions, and functions). Then, we fine-tune the open-source LLM with 7B parameters on Repo-Instruct to produce a strong code completion baseline model Qwen2.5-Coder-Instruct-C based on the open-source model. Qwen2.5-Coder-Instruct-C is rigorously evaluated against existing benchmarks, including MultiPL-E and ExecRepoBench, which consistently outperforms prior baselines across all programming languages. The deployment of \ourmethod{} can be used as a high-performance, local service for programming development\footnote{\url{https://execrepobench.github.io/}}.

Pattern images are everywhere in the digital and physical worlds, and tools to edit them are valuable. But editing pattern images is tricky: desired edits are often programmatic: structure-aware edits that alter the underlying program which generates the pattern. One could attempt to infer this underlying program, but current methods for doing so struggle with complex images and produce unorganized programs that make editing tedious. In this work, we introduce a novel approach to perform programmatic edits on pattern images. By using a pattern analogy -- a pair of simple patterns to demonstrate the intended edit -- and a learning-based generative model to execute these edits, our method allows users to intuitively edit patterns. To enable this paradigm, we introduce SplitWeave, a domain-specific language that, combined with a framework for sampling synthetic pattern analogies, enables the creation of a large, high-quality synthetic training dataset. We also present TriFuser, a Latent Diffusion Model (LDM) designed to overcome critical issues that arise when naively deploying LDMs to this task. Extensive experiments on real-world, artist-sourced patterns reveals that our method faithfully performs the demonstrated edit while also generalizing to related pattern styles beyond its training distribution.

Recently, it is often said that the data used for the pre-training of large language models (LLMs) have been exhausted. This paper proposes a solution to the problem: Automated generation of massive reasonable empirical theorems by forward reasoning based on strong relevant logics. In fact, this can be regarded as a part of our approach to the problems of ATF (Automated Theorem Finding) and AKA (Automated Knowledge Appreciation).

We improve unlearning efficacy by introducing random labeling loss and ensuring the model retains its general-purpose knowledge by adjusting targeted parameters with gradient-based weight saliency.