AI Systems for Agents, Science, and Medicine.

Deep Research Agents generate analyst-grade reports, yet evaluating them remains challenging due to the absence of a single ground truth and the multidimensional nature of research quality. Recent benchmarks propose distinct methodologies, yet they suffer from the Mirage of Synthesis, where strong surface-level fluency and citation alignment can obscure underlying factual and reasoning defects. To address this, we propose DREAM (Deep Research Evaluation with Agentic Metrics), a framework that instantiates the principle of capability parity by making evaluation itself agentic. DREAM structures assessment through an evaluation protocol combining query-agnostic metrics with adaptive metrics generated by a tool-calling agent, enabling temporally aware coverage, grounded verification, and systematic reasoning probes. Controlled evaluations demonstrate DREAM is significantly more sensitive to factual and temporal decay than existing benchmarks, offering a scalable, reference-free evaluation paradigm.

Large language models (LLMs) often exhibit performance disparities across languages, with naive multilingual fine-tuning frequently degrading performance due to negative cross-lingual interference. To address this, we introduce COMPASS (COntinual Multilingual PEFT with Adaptive Semantic Sampling), a novel data-centric framework for adapting LLMs to target languages. COMPASS leverages parameter-efficient fine-tuning (PEFT) by training lightweight, language-specific adapters on a judiciously selected subset of auxiliary multilingual data. The core of our method is a distribution-aware sampling strategy that uses multilingual embeddings and clustering to identify semantic gaps between existing training data and a target usage distribution. By prioritizing auxiliary data from under-represented semantic clusters, COMPASS maximizes positive cross-lingual transfer while minimizing interference. We extend this into a continual learning framework, COMPASS-ECDA, which monitors for data distribution shifts in production and dynamically updates adapters to prevent model staleness, balancing adaptation to new data with the preservation of existing knowledge. Across three different model architectures (Phi-4-Mini, Llama-3.1-8B, and Qwen2.5-7B) and multiple challenging multilingual benchmarks (Global-MMLU, MMLUProX), including unseen long-context tasks (OneRuler), we demonstrate that COMPASS consistently outperforms baseline methods guided by linguistic similarity, providing an effective, efficient, and sustainable solution for developing and maintaining high-performing multilingual models in dynamic environments.

We present Amazon Nova, a new generation of state-of-the-art foundation models that deliver frontier intelligence and industry-leading price performance. Amazon Nova Pro is a highly-capable multimodal model with the best combination of accuracy, speed, and cost for a wide range of tasks. Amazon Nova Lite is a low-cost multimodal model that is lightning fast for processing images, video, documents and text. Amazon Nova Micro is a text-only model that delivers our lowest-latency responses at very low cost. Amazon Nova Canvas is an image generation model that creates professional grade images with rich customization controls. Amazon Nova Reel is a video generation model offering high-quality outputs, customization, and motion control. Our models were built responsibly and with a commitment to customer trust, security, and reliability. We report benchmarking results for core capabilities, agentic performance, long context, functional adaptation, runtime performance, and human evaluation.

Southeast Asia (SEA) is a region rich in linguistic diversity and cultural variety, with over 1,300 indigenous languages and a population of 671 million people. However, prevailing AI models suffer from a significant lack of representation of texts, images, and audio datasets from SEA, compromising the quality of AI models for SEA languages. Evaluating models for SEA languages is challenging due to the scarcity of high-quality datasets, compounded by the dominance of English training data, raising concerns about potential cultural misrepresentation. To address these challenges, we introduce SEACrowd, a collaborative initiative that consolidates a comprehensive resource hub that fills the resource gap by providing standardized corpora in nearly 1,000 SEA languages across three modalities. Through our SEACrowd benchmarks, we assess the quality of AI models on 36 indigenous languages across 13 tasks, offering valuable insights into the current AI landscape in SEA. Furthermore, we propose strategies to facilitate greater AI advancements, maximizing potential utility and resource equity for the future of AI in SEA.

Cytochrome P450 enzymes aid in the elimination of a preponderance of small molecule drugs, but can generate reactive metabolites that may adversely react with protein and DNA and prompt drug candidate attrition or market withdrawal. We previously introduced XenoNet, a method to enumerate and predict likely metabolic pathways and intermediate metabolites formed during drug metabolism. Here, we extend XenoNet with a bidirectional message passing neural network that integrates edge feature information and local network structure using edge-conditioned graph convolutions and jumping knowledge to predict the authenticity of inferred Phase I metabolite structures. We test on 311 metabolic networks, containing 6606 intermediates, annotated from 20 736 reaction records derived from chemical literature and Cytochrome P450 databases. Our model achieved area under the receiver operating characteristic curves of 88.5 and 87.6% for separating experimentally observed and unobserved metabolites at global and network levels, respectively. Our approach is robust across varying network sizes, accurately captures metabolites that appear or disappear depending on context, extracts important metabolic subnetworks, and identifies known bioactivation pathways for drugs like nimesulide and terbinafine. Our method can accurately suggest unreported metabolites for experimental study and may rationalize modifications for avoiding deleterious pathways antecedent to reactive metabolite formation.

Drug-drug interactions account for up to 30% of adverse drug reactions. Increasing prevalence of electronic health records (EHRs) offers a unique opportunity to build machine learning algorithms to identify drug-drug interactions that drive adverse events. In this study, we investigated hospitalizations’ data to study drug interactions with non-steroidal anti-inflammatory drugs (NSAIDS) that result in drug-induced liver injury (DILI). We propose a logistic regression based machine learning algorithm that unearths several known interactions from an EHR dataset of about 400,000 hospitalization. Our proposed modeling framework is successful in detecting 87.5% of the positive controls, which are defined by drugs known to interact with diclofenac causing an increased risk of DILI, and correctly ranks aggregate risk of DILI for eight commonly prescribed NSAIDs. We found that our modeling framework is particularly successful in inferring associations of drug-drug interactions from relatively small EHR datasets. Furthermore, we have identified a novel and potentially hepatotoxic interaction that might occur during concomitant use of meloxicam and esomeprazole, which are commonly prescribed together to allay NSAID-induced gastrointestinal (GI) bleeding. Empirically, we validate our approach against prior methods for signal detection on EHR datasets, in which our proposed approach outperforms all the compared methods across most metrics, such as area under the receiver operating characteristic curve (AUROC) and area under the precision-recall curve (AUPRC).

Drug metabolism is a common cause of adverse drug reactions. Drug molecules can be metabolized into reactive metabolites, which can conjugate to biomolecules, like protein and DNA, in a process termed bioactivation. Cytochrome P450 (P450) enzymes metabolize drugs and generate reactive metabolites. However, traditional experimental and computational methods are insufficient to uncover all of the possible intermediate metabolites that form during drug metabolism. We developed XenoNet, a metabolic network predictor that takes a pair of a substrate and target product as input and enumerates pathways, or sequences of intermediate metabolite structures, between the pair. XenoNet then computes the likelihood of those pathways and intermediate metabolites. XenoNet was trained on all publicly available metabolic networks constructed from metabolite structures. We validated XenoNet against 17 054 metabolic networks and found XenoNet can predict experimentally observed pathways and intermediate metabolites linking the input substrate and product pair with a recall of 88 and 46%, respectively. With likelihood scoring, we achieved a top-one pathway and intermediate metabolite accuracy of 93.6 and 51.9%, respectively, and outperformed popular prediction methods including BioTransformer and SyGMa. XenoNet is available at https://swami.wustl.edu/xenonet.