Do LLMs Know What to Ask and When? Evaluating Multi-Turn Information Seeking

When a user question is underspecified, a capable model should recognize that its context is insufficient, identify the missing information, ask for it, and respond only once that information determines a unique answer. We formalize multi-turn information seeking as solving a k-underspecified constraint satisfaction problem, where k is the number of variables jointly required to determine the target and therefore measures the degree of missing information. We instantiate the formulation in MT-InfoSeek, a controlled evaluation suite of 5,251 problems and 9,006 task instances spanning mathematics, logic, biology, medicine, and general knowledge. We evaluate models along three axes: what they ask, when they ask it, and how the acquired information affects the final answer. Performance degrades across models and domains as underspecification increases. Models recognize that additional information is needed but underestimate how much, and in logical problems at k = 2 they under-predict the degree of missing information about four times as often as they over-predict it. They also fail to identify a minimal sufficient set of queries, improve only marginally when given the true k, and often stop before acquiring sufficient information. In tasks with ordered dependencies, an incorrect query order reduces final accuracy even when the model eventually acquires all necessary information. We measure information seeking directly through final sufficiency, which records whether the acquired information determines the target independent of answer generation. This separation shows differences between models that final accuracy alone does not capture, and indicates that the ability to seek information over multiple turns is distinct from the ability to generate answers and is not measured by current LLM evaluations.

MatrAIx: Simulating the World with 8.3 Billion Persona Agents

Human evaluation of AI systems and digital products is costly, slow, and difficult to scale. Offline evaluations are more scalable but often abstract away human diversity and interactive behavior. We therefore introduce MatrAIx, a population-scale simulated-user evaluation infrastructure for testing AI systems and digital products with heterogeneous users. MatrAIx has three core components: First, Persona 8B contains 8.3 billion persona records represented by 1,290 categorical dimensions. Records are either sampled from a dependency graph that preserves correlated attributes or derived from human-authored profiles. We release a quality-filtered coreset of approximately 1 million personas, comprising 599,847 human-grounded and 400,000 synthetic records. Second, the MatrAIx Playground provides four environments in which diverse users evaluate and interact with digital products: Survey, AI Chatbot, Web, and App. Third, MatrAIx provides 1,010 application tasks spanning more than 25 domains, including Commerce, Software, Finance, and Healthcare. We conducted 18,189 evaluation trials across eight representative tasks. Persona agents were powered by three LLMs: Claude Opus 4.8, GPT 5.5, and Claude Haiku 4.5. The resulting feedback captures how decisions and preferences vary across persona backgrounds, including hesitation after a price increase, willingness to continue after an AI assistant fails, and latency tolerance. We conducted two main validation studies: First, a 400-trial controlled study evaluated persona adherence across ten behavioral attributes and all four environments. The declared behavior was expressed or correctly suppressed in 366 trials (91.5%). Second, human and LLM judges evaluated the extraction quality of human-grounded personas. Overall, MatrAIx provides an end-to-end infrastructure for evaluating AI systems and digital products with diverse simulated human users.

An AI agent for treatment reasoning over a biomedical tool universe

Treatment reasoning underpins every therapeutic decision, integrating disease context, comorbidities, medications, contraindications, and evolving biomedical knowledge to select an appropriate therapy. It is inherently iterative: candidates are weighed against many constraints, revised as evidence emerges, and grounded in verifiable sources. Here we introduce ATHENA-R1, an AI agent for treatment reasoning across all FDA approved drugs since 1939, trained by reinforcement learning over a universe of 212 biomedical tools. At each step it identifies missing information, selects and runs relevant tools, and incorporates the evidence. To train it without human-annotated traces, we build a two-level self-learning framework: multi-agent systems construct the tools, tasks, and reasoning trajectories for supervised fine-tuning, then reinforcement learning with scientific feedback rewards reasoning quality (evidence gathering, grounded tool use, logical non-redundancy). Across five benchmarks of 3,168 drug reasoning tasks and 456 patient treatment cases, ATHENA-R1 outperforms language models and tool-use systems, reaching 94.7% accuracy on open-ended drug reasoning and 82.9% on treatment reasoning, 17.8 and 10.7 points above GPT-5. In blinded evaluations by experts from 28 rare disease organizations, it is preferred over reference models on all criteria, and physicians rated it favorably on complex hospitalized cardiovascular and infectious-disease cases. Adverse-event hypotheses it generated, tested in electronic health records from 5.4 million patients, reached adjusted odds ratios of 1.48-1.84, with no elevation among negative controls. Because it requires knowing what evidence to seek before concluding, treatment reasoning has long been hard for AI; we show it can be reframed as a learnable process of iterative evidence gathering that reinforcement learning can train AI to perform.

How Post-Training Shapes Biological Reasoning Models

Scientific reasoning models for biology combine language models with foundation models trained on multimodal biological data, including DNA, RNA, and proteins. These models are built through post-training, yet how each stage shapes reasoning and generalization remains poorly understood. We study when post-training improves performance and when it induces over-specialization. Across genomics, transcriptomics, and proteins, we train and evaluate more than 100 biological reasoning models under controlled variation in backbone, continued pre-training (CPT), supervised fine-tuning (SFT), and reinforcement learning (RL), measuring both in-domain (ID) and out-of-domain (OOD) performance. We find that each post-training stage reshapes generalization in a distinct way rather than contributing uniform gains. CPT improves downstream performance by aligning models with biological language. SFT consistently increases ID performance but causes OOD performance to peak early and decline as models fit the training distribution. RL, when applied to strong SFT checkpoints with aligned rewards, improves OOD performance and partially recovers generalization. These results show that biological reasoning does not improve monotonically with additional supervision or compute. Instead, performance depends on how training stages are composed. Under fixed post-training budgets, the strongest ID-OOD trade-off comes from brief SFT, larger RL allocations, and asymmetric adaptation capacity across stages.

Benchmarking AI Agents for Addressing Scientific Challenges Across Scales

AI agents are increasingly being developed to accelerate scientific discovery, yet their practical capabilities in real research settings remain poorly understood. Existing benchmarks for AI agents rarely capture the complexity, heterogeneity, and extended reasoning required by scientific work, whereas benchmarks for scientific tasks often reduce research to static, direct problems and provide limited support for interactive evaluation. Here, we introduce SciAgentArena, a systematic benchmark for evaluating AI agents in real-world scientific research scenarios drawn from emerging needs across multiple domains. SciAgentArena comprises approximately 200 tasks with stepwise verification and an interactive, agent-agnostic environment for assessing diverse AI agents. Using this benchmark, we find that current agents can contribute effectively to well-specified data-analysis workflows, particularly when the task structure and evaluation criteria are clear. However, their performance remains uneven across scientific contexts: agents struggle to generate genuinely novel insights, sustain self-directed exploration, and formulate robust solutions for open-ended research questions. We further characterize common failure modes across agents and identify opportunities for improving their reliability, autonomy, and scientific reasoning. Together, SciAgentArena provides a practical framework for measuring progress in AI agents for science and for guiding the design of future agents capable of addressing complex scientific challenges. Full codes, tasks, and datasets can be accessed via this link: https://sciagentarena.github.io/.

AutoScientists: Self-Organizing Agent Teams for Long-Running Scientific Experimentation

Scientific research proceeds through iterative cycles of hypothesis generation, experiment design, execution, and revision. AI agents can automate parts of this process, but existing approaches typically follow a single research trajectory or coordinate through a central planner with fixed objectives. As a result, they struggle to sustain parallel exploration, adapt as experimental evidence changes, or preserve knowledge of failed directions over long-running experiments. We introduce AutoScientists, a decentralized team of AI agents for long-running computational scientific experimentation. Agents interpret a shared experimental state, self-organize into teams around promising hypotheses, critique proposals before using experimental compute, and share successes and failures to reduce redundant exploration. Under matched experimental budgets, AutoScientists improves over prior AI agents across biomedical machine learning, language-model training optimization, and protein fitness prediction. On BioML-Bench, spanning biomedical imaging, protein engineering, single-cell omics, and drug discovery, AutoScientists achieves a mean leaderboard percentile of 74.4% across 24 tasks, improving over the strongest AI agent by +8.33%. On GPT training optimization, AutoScientists reaches a target validation bits-per-byte 1.9x faster than Autoresearch and continues discovering improvements from a starting champion where the single-agent approach finds none (7 vs. 0 accepted improvements). On ProteinGym fitness prediction, AutoScientists discovers a method for ACE2-Spike binding that improves over the current state-of-the-art model by +12.5% in Spearman correlation. Applied without modification across all 217 ProteinGym assays, the same method improves over the prior state of the art by +6.5% (Spearman correlation).

PhylaFlow: Hybrid Flow Matching in Billera-Holmes-Vogtmann Tree Space for Phylogenetic Inference

Phylogenetic trees are hybrid objects: branch lengths vary continuously, while topologies change discretely through edge contractions and expansions. Billera-Holmes-Vogtmann (BHV) tree space provides a canonical geometry for this structure, representing each resolved topology as a Euclidean orthant and topological changes as motion across shared lower-dimensional boundaries. We introduce PhylaFlow, a hybrid flow-matching model that learns posterior-basin transport in BHV tree space. PhylaFlow is trained on BHV geodesic paths from random starting trees to short-run posterior samples, coupling continuous branch-length motion within orthants with learned boundary events and discrete topology transitions. We evaluate the learned geometry operationally: if the flow reaches posterior-relevant regions, finite-budget Bayesian refinement initialized from, or guided by, its terminal trees should recover posterior-supported topologies more efficiently. Across DS1-DS8 phylogenetic posterior benchmarks, PhylaFlow substantially reduces initial Tree-KL relative to classical initializers. After finite-budget MrBayes refinement, direct PhylaFlow improves early and intermediate topology-recovery trajectories on most datasets, while split-guided PhylaFlow-MCMC obtains the strongest hard-case results. The best PhylaFlow variant outperforms short-warmup on seven of eight datasets and PhyloGFN on five of eight under the same refinement budget. In a joint sequence-conditioned experiment, sequence embeddings steer posterior split recovery, although exact posterior topology recovery remains preliminary. These results show that hybrid flow matching can learn actionable transport in BHV tree space and provide a geometry-aware proposal mechanism for Bayesian phylogenetic inference.

Evaluating Relational Reasoning in LLMs with REL

Relational reasoning is the ability to infer relations that jointly bind multiple entities, attributes, or variables. This ability is central to scientific reasoning, but existing evaluations of relational reasoning in large language models often focus on structured inputs such as tables, graphs, or synthetic tasks, and do not isolate the difficulty introduced by higher-arity relational binding. We study this problem through the lens of Relational Complexity (RC), which we define as the minimum number of independent entities or operands that must be simultaneously bound to apply a relation. RC provides a principled way to vary reasoning difficulty while controlling for confounders such as input size, vocabulary, and representational choices. Building on RC, we introduce REL, a generative benchmark framework spanning algebra, chemistry, and biology that varies RC within each domain. Across frontier LLMs, performance degrades consistently and monotonically as RC increases, even when the total number of entities is held fixed. This failure mode persists with increased test-time compute and in-context learning, suggesting a limitation tied to the arity of the required relational binding rather than to insufficient inference steps or lack of exposure to examples. Our results identify a regime of higher-arity reasoning in which current models struggle, and motivate re-examining benchmarks through the lens of relational complexity.

Distribution-Conditioned Transport

Learning a transport model that maps a source distribution to a target distribution is a canonical problem in machine learning, but scientific applications increasingly require models that can generalize to source and target distributions unseen during training. We introduce distribution-conditioned transport (DCT), a framework that conditions transport maps on learned embeddings of source and target distributions, enabling generalization to unseen distribution pairs. DCT also allows semi-supervised learning for distributional forecasting problems: because it learns from arbitrary distribution pairs, it can leverage distributions observed at only one condition to improve transport prediction. DCT is agnostic to the underlying transport mechanism, supporting models ranging from flow matching to distributional divergence-based models (e.g. Wasserstein, MMD). We demonstrate the practical performance benefits of DCT on synthetic benchmarks and four applications in biology: batch effect transfer in single-cell genomics, perturbation prediction from mass cytometry data, learning clonal transcriptional dynamics in hematopoiesis, and modeling T-cell receptor sequence evolution.

When Sensors Fail: Temporal Sequence Models for Robust PPO under Sensor Drift

Real-world reinforcement learning systems must operate under distributional drift in their observation streams, yet most policy architectures implicitly assume fully observed and noise-free states. We study robustness of Proximal Policy Optimization (PPO) under temporally persistent sensor failures that induce partial observability and representation shift. To respond to this drift, we augment PPO with temporal sequence models, including Transformers and State Space Models (SSMs), to enable policies to infer missing information from history and maintain performance. Under a stochastic sensor failure process, we prove a high-probability bound on infinite-horizon reward degradation that quantifies how robustness depends on policy smoothness and failure persistence. Empirically, on MuJoCo continuous-control benchmarks with severe sensor dropout, we show Transformer-based sequence policies substantially outperform MLP, RNN, and SSM baselines in robustness, maintaining high returns even when large fractions of sensors are unavailable. These results demonstrate that temporal sequence reasoning provides a principled and practical mechanism for reliable operation under observation drift caused by sensor unreliability.