Tool-calling is essential for Large Language Model (LLM) agents to complete real-world tasks. While most existing benchmarks assume simple, perfectly documented tools, real-world tools (e.g., general "search" APIs) are often opaque, lacking clear best practices or failure modes. Can LLM agents improve their performance in environments with opaque tools by interacting and subsequently improving documentation? To study this, we create OpaqueToolsBench, a benchmark consisting of three distinct task-oriented environments: general function calling, interactive chess playing, and long-trajectory agentic search. Each environment provides underspecified tools that models must learn to use effectively to complete the task. Results on OpaqueToolsBench suggest existing methods for automatically documenting tools are expensive and unreliable when tools are opaque. To address this, we propose a simple framework, ToolObserver, that iteratively refines tool documentation by observing execution feedback from tool-calling trajectories. Our approach outperforms existing methods on OpaqueToolsBench across datasets, even in relatively hard settings. Furthermore, for test-time tool exploration settings, our method is also efficient, consuming 3.5-7.5x fewer total tokens than the best baseline.

We present HealMed, an expert-reviewed benchmark for multilingual evaluation of large language models in medicine. HealMed contains 1,000 examples in each of nine languages, drawn from nine datasets and covering three task formats: MCQA, NLI and open-ended QA. The benchmark was developed over two years by 23 physicians and medical experts based across nine countries and regions. Each translation was evaluated and revised by two experts fluent in English and the corresponding target language. On HealMed, performance declined most in low-resource languages, although the size of the gap varied markedly across languages and models. The strongest proprietary models were the most stable across languages, whereas many open-source and medically specialized models showed larger and less consistent gaps. Medical specialization alone did not ensure multilingual robustness. Furthermore, expert revision could either raise or lower measured performance, indicating that translation quality materially affects cross-language evaluation results.

Modern deep neural network (DNN) applications integrate multiple DNN models into inference pipelines with stringent latency requirements for customized tasks. To mitigate extensive request timeouts caused by accumulation, systems for inference pipelines commonly drop a subset of requests so the remaining ones can satisfy latency constraints. Since it is commonly believed that request dropping adversely affects goodput, existing systems only drop requests when they have to, which we call reactive dropping. However, this reactive policy can not maintain high goodput, as it neither makes timely dropping decisions nor identifies the proper set of requests to drop, leading to issues of dropping requests too late or dropping the wrong set of requests. We propose that the inference system should proactively drop certain requests in advance to enhance the goodput across the entire workload. To achieve this, we design an inference system PARD. It enhances goodput with timely and precise dropping decisions by integrating a proactive dropping method that decides when to drop requests using runtime information of the inference pipeline, and an adaptive request priority mechanism that selects which specific requests to drop based on remaining latency budgets and workload intensity. Evaluation on a cluster of 64 GPUs over real-world workloads shows that PARD achieves 16%-176% higher goodput than the state of the art while reducing the drop rate and wasted computation resources by 1.6×-17× and 1.5×-62× respectively.

AI agents are increasingly deployed for professional investment research, yet no benchmark captures the complexity of the full investor workflow. Existing benchmarks mainly target financial data extraction, a narrow slice that current models have largely saturated, while reference-based metrics and generic LLM-as-a-judge scoring fall short on the open-ended, long-form answers that real analyst queries demand. We introduce FrontierFinance, a fully open benchmark of 220 expert-crafted queries and 11,543 source-attributed rubrics spanning six crucial use cases across the full investor workflow. FrontierFinance is both broader and harder than existing public finance benchmarks. Evaluating frontier models and agent systems under a common harness restricted to publicly available data, we find that the tool harness, not the model alone, strongly shapes quality and efficiency; that Samaya's in-house system leads at 56.0

Vision-Language Models (VLMs) have demonstrated strong performance on textbook-style physics problems, yet they frequently fail when confronted with dynamic real-world scenarios that require temporal consistency and causal reasoning across frames. We identify two fundamental challenges underlying these failures: (1) spatio-temporal identity drift, where objects lose their physical identity across successive frames and break causal chains, and (2) volatility of inference-time insights, where a model may occasionally produce correct physical reasoning but never consolidates it for future reuse. To address these challenges, we propose PhysNote, an agentic framework that enables VLMs to externalize and refine physical knowledge through self-generated "Knowledge Notes." PhysNote stabilizes dynamic perception through spatio-temporal canonicalization, organizes self-generated insights into a hierarchical knowledge repository, and drives an iterative reasoning loop that grounds hypotheses in visual evidence before consolidating verified knowledge. Experiments on PhysBench demonstrate that PhysNote achieves 56.68

Long-horizon agentic search requires iteratively exploring the web over long trajectories and synthesizing information across many sources, enabling powerful applications like deep research systems. In this work, we show that popular agentic search frameworks struggle to scale to long trajectories primarily due to context limitations–they accumulate long, noisy content, hit context window and tool budgets, or stop early. We therefore introduce SLIM (Simple Lightweight Information Management), a simple framework that separates retrieval into distinct search and browse tools, and periodically summarizes the trajectory, keeping context concise while enabling longer, more focused searches. Across a wide range of long-horizon tasks, SLIM achieves comparable performance at substantially lower cost and far fewer tool calls than strong open-source frameworks with both proprietary and open-weight models, including RL-trained models for deep research. Specifically, with o3 as the base model, SLIM achieves 56

Instruction-tuned language models (LM) are able to respond to imperative commands, providing a more natural user interface compared to their base counterparts. In this work, we present Promptriever, the first retrieval model able to be prompted like an LM. To train Promptriever, we curate and release a new instance-level instruction training set from MS MARCO, spanning nearly 500k instances. Promptriever not only achieves strong performance on standard retrieval tasks, but also follows instructions. We observe: (1) large gains (reaching SoTA) on following detailed relevance instructions (+14.3 p-MRR / +3.1 nDCG on FollowIR), (2) significantly increased robustness to lexical choices/phrasing in the query+instruction (+12.9 Robustness@10 on InstructIR), and (3) the ability to perform hyperparameter search via prompting to reliably improve retrieval performance (+1.4 average increase on BEIR). Promptriever demonstrates that retrieval models can be controlled with prompts on a per-query basis, setting the stage for future work aligning LM prompting techniques with information retrieval.

Question answering based on retrieval augmented generation (RAG-QA) is an important research topic in NLP and has a wide range of real-world applications. However, most existing datasets for this task are either constructed using a single source corpus or consist of short extractive answers, which fall short of evaluating large language model (LLM) based RAG-QA systems on cross-domain generalization. To address these limitations, we create Long-form RobustQA (LFRQA), a new dataset comprising human-written long-form answers that integrate short extractive answers from multiple documents into a single, coherent narrative, covering 26K queries and large corpora across seven different domains. We further propose RAG-QA Arena by directly comparing model-generated answers against LFRQA’s answers using LLMs as evaluators. We show via extensive experiments that RAG-QA Arena and human judgments on answer quality are highly correlated. Moreover, only 41.3% of the most competitive LLM’s answers are preferred to LFRQA’s answers, demonstrating RAG-QA Arena as a challenging evaluation platform for future research.

Modern language models (LMs) need to follow human instructions while being faithful; yet, they often fail to achieve both. Here, we provide concrete evidence of a trade-off between instruction following (i.e., follow open-ended instructions) and faithfulness (i.e., ground responses in given context) when training LMs with these objectives. For instance, fine-tuning LLaMA-7B on instruction following datasets renders it less faithful. Conversely, instruction-tuned Vicuna-7B shows degraded performance at following instructions when further optimized on tasks that require contextual grounding. One common remedy is multi-task learning (MTL) with data mixing, yet it remains far from achieving a synergic outcome. We propose a simple yet effective method that relies on Rejection Sampling for Continued Self-instruction Tuning (ReSet), which significantly outperforms vanilla MTL. Surprisingly, we find that less is more, as training ReSet with high-quality, yet substantially smaller data (three-fold less) yields superior results. Our findings offer a better understanding of objective discrepancies in alignment training of LMs.

The open-ended Visual Question Answering (VQA) task requires AI models to jointly reason over visual and natural language inputs using world knowledge. Recently, pre-trained Language Models (PLM) such as GPT-3 have been applied to the task and shown to be powerful world knowledge sources. However, these methods suffer from low knowledge coverage caused by PLM bias -- the tendency to generate certain tokens over other tokens regardless of prompt changes, and high dependency on the PLM quality -- only models using GPT-3 can achieve the best result. To address the aforementioned challenges, we propose RASO: a new VQA pipeline that deploys a generate-then-select strategy guided by world knowledge for the first time. Rather than following the de facto standard to train a multi-modal model that directly generates the VQA answer, RASO first adopts PLM to generate all the possible answers, and then trains a lightweight answer selection model for the correct answer. As proved in our analysis, RASO expands the knowledge coverage from in-domain training data by a large margin. We provide extensive experimentation and show the effectiveness of our pipeline by advancing the state-of-the-art by 4.1% on OK-VQA, without additional computation cost. Code and models are released at http://cogcomp.org/page/publication_view/1010

Generative models have been widely applied to solve extractive tasks, where parts of the input is extracted to form the desired output, and achieved significant success. For example, in extractive question answering (QA), generative models have constantly yielded state-of-the-art results. In this work, we identify the issue of tokenization inconsistency that is commonly neglected in training these models. This issue damages the extractive nature of these tasks after the input and output are tokenized inconsistently by the tokenizer, and thus leads to performance drop as well as hallucination. We propose a simple yet effective fix to this issue and conduct a case study on extractive QA. We show that, with consistent tokenization, the model performs better in both in-domain and out-of-domain datasets, with a notable average of +1.7 F2 gain when a BART model is trained on SQuAD and evaluated on 8 QA datasets. Further, the model converges faster, and becomes less likely to generate out-of-context answers. With these findings, we would like to call for more attention on how tokenization should be done when solving extractive tasks and recommend applying consistent tokenization during training.

Open-domain question answering (ODQA) is a crucial task in natural language processing. A typical ODQA system relies on a retriever module to select relevant contexts from a large corpus for a downstream reading comprehension model. Existing ODQA datasets consist mainly of Wikipedia corpus, and are insufficient to study models’ generalizability across diverse domains, as models are trained and evaluated on the same genre of data. We propose RobustQA 1 , a novel benchmark consisting of datasets from 8 different domains, which facilitates the evaluation of ODQA’s domain robustness. To build RobustQA , we annotate QA pairs in retrieval datasets with rigorous quality control. We further examine improving QA performances by incorporating unsupervised learning methods with target-domain corpus and adopting large generative language models. These methods can effectively improve model performances on RobustQA . However, experimental results demonstrate a significant gap from in-domain training, suggesting that Ro-bustQA is a challenging benchmark to evaluate ODQA domain robustness.

There has been great progress in unifying various table-to-text tasks using a single encoder-decoder model trained via multi-task learning (Xie et al., 2022). However, existing methods typically encode task information with a simple dataset name as a prefix to the encoder. This not only limits the effectiveness of multi-task learning, but also hinders the model's ability to generalize to new domains or tasks that were not seen during training, which is crucial for real-world applications. In this paper, we propose compositional task configurations, a set of prompts prepended to the encoder to improve cross-task generalization of unified models. We design the task configurations to explicitly specify the task type, as well as its input and output types. We show that this not only allows the model to better learn shared knowledge across different tasks at training, but also allows us to control the model by composing new configurations that apply novel input-output combinations in a zero-shot manner. We demonstrate via experiments over ten table-to-text tasks that our method outperforms the UnifiedSKG baseline by noticeable margins in both in-domain and zero-shot settings, with average improvements of +0.5 and +12.6 from using a T5-large backbone, respectively.

We introduce the initial release of our software Robustar, which aims to improve the robustness of vision classification machine learning models through a data-driven perspective. Building upon the recent understanding that the lack of machine learning model's robustness is the tendency of the model's learning of spurious features, we aim to solve this problem from its root at the data perspective by removing the spurious features from the data before training. In particular, we introduce a software that helps the users to better prepare the data for training image classification models by allowing the users to annotate the spurious features at the pixel level of images. To facilitate this process, our software also leverages recent advances to help identify potential images and pixels worthy of attention and to continue the training with newly annotated data. Our software is hosted at the GitHub Repository https://github.com/HaohanWang/Robustar.

Much recent work suggests that incorporating syntax information from dependency trees can improve task-specific transformer models. However, the effect of incorporating dependency tree information into pre-trained transformer models (e.g., BERT) remains unclear, especially given recent studies highlighting how these models implicitly encode syntax. In this work, we systematically study the utility of incorporating dependency trees into pre-trained transformers on three representative information extraction tasks: semantic role labeling (SRL), named entity recognition, and relation extraction. We propose and investigate two distinct strategies for incorporating dependency structure: a late fusion approach, which applies a graph neural network on the output of a transformer, and a joint fusion approach, which infuses syntax structure into the transformer attention layers. These strategies are representative of prior work, but we introduce additional model design elements that are necessary for obtaining improved performance. Our empirical analysis demonstrates that these syntax-infused transformers obtain state-of-the-art results on SRL and relation extraction tasks. However, our analysis also reveals a critical shortcoming of these models: we find that their performance gains are highly contingent on the availability of human-annotated dependency parses, which raises important questions regarding the viability of syntax-augmented transformers in real-world applications.

The MEDIQA 2021 shared tasks at the BioNLP 2021 workshop addressed three tasks on summarization for medical text: (i) a question summarization task aimed at exploring new approaches to understanding complex real-world consumer health queries, (ii) a multi-answer summarization task that targeted aggregation of multiple relevant answers to a biomedical question into one concise and relevant answer, and (iii) a radiology report summarization task addressing the development of clinically relevant impressions from radiology report findings. Thirty-five teams participated in these shared tasks with sixteen working notes submitted (fifteen accepted) describing a wide variety of models developed and tested on the shared and external datasets. In this paper, we describe the tasks, the datasets, the models and techniques developed by various teams, the results of the evaluation, and a study of correlations among various summarization evaluation measures. We hope that these shared tasks will bring new research and insights in biomedical text summarization and evaluation.

We introduce biomedical and clinical English model packages for the Stanza Python NLP library. These packages offer accurate syntactic analysis and named entity recognition capabilities for biomedical and clinical text, by combining Stanza's fully neural architecture with a wide variety of open datasets as well as large-scale unsupervised biomedical and clinical text data. We show via extensive experiments that our packages achieve syntactic analysis and named entity recognition performance that is on par with or surpasses state-of-the-art results. We further show that these models do not compromise speed compared to existing toolkits when GPU acceleration is available, and are made easy to download and use with Stanza's Python interface. A demonstration of our packages is available at: http://stanza.run/bio.

Learning visual representations of medical images (e.g., X-rays) is core to medical image understanding but its progress has been held back by the scarcity of human annotations. Existing work commonly relies on fine-tuning weights transferred from ImageNet pretraining, which is suboptimal due to drastically different image characteristics, or rule-based label extraction from the textual report data paired with medical images, which is inaccurate and hard to generalize. Meanwhile, several recent studies show exciting results from unsupervised contrastive learning from natural images, but we find these methods help little on medical images because of their high inter-class similarity. We propose ConVIRT, an alternative unsupervised strategy to learn medical visual representations by exploiting naturally occurring paired descriptive text. Our new method of pretraining medical image encoders with the paired text data via a bidirectional contrastive objective between the two modalities is domain-agnostic, and requires no additional expert input. We test ConVIRT by transferring our pretrained weights to 4 medical image classification tasks and 2 zero-shot retrieval tasks, and show that it leads to image representations that considerably outperform strong baselines in most settings. Notably, in all 4 classification tasks, our method requires only 10% as much labeled training data as an ImageNet initialized counterpart to achieve better or comparable performance, demonstrating superior data efficiency.

Extracting structured clinical information from free-text radiology reports can enable the use of radiology report information for a variety of critical healthcare applications. In our work, we present RadGraph, a dataset of entities and relations in full-text chest X-ray radiology reports based on a novel information extraction schema we designed to structure radiology reports. We release a development dataset, which contains board-certified radiologist annotations for 500 radiology reports from the MIMIC-CXR dataset (14,579 entities and 10,889 relations), and a test dataset, which contains two independent sets of board-certified radiologist annotations for 100 radiology reports split equally across the MIMIC-CXR and CheXpert datasets. Using these datasets, we train and test a deep learning model, RadGraph Benchmark, that achieves a micro F1 of 0.82 and 0.73 on relation extraction on the MIMIC-CXR and CheXpert test sets respectively. Additionally, we release an inference dataset, which contains annotations automatically generated by RadGraph Benchmark across 220,763 MIMIC-CXR reports (around 6 million entities and 4 million relations) and 500 CheXpert reports (13,783 entities and 9,908 relations) with mappings to associated chest radiographs. Our freely available dataset can facilitate a wide range of research in medical natural language processing, as well as computer vision and multi-modal learning when linked to chest radiographs.