Tracked across scientific fields
A chronological view of public research related to this topic.
Related updates
[This is a guest post by Sophie Huiberts , being an expanded version of a post on her own blog . This blog post was initially written in a different file format and converted using AI. — T.] When you prove a theorem, you are imparted a responsibility to the field of mathematics. At minimum you need to explain what you did, including which steps required new ideas and which steps are already well-known. More substanti
Welcome to Import AI, a newsletter about AI research. Import AI runs on arXiv, cappuccinos, and feedback from readers. If you’d like to support this, please subscribe. Subscribe now New benchmark shows how AI models lack intuitive visual reasoning: …Humanity’s Sixth Sense highlights some shortcomings of current AI systems… AI startups Elorian and Scale AI have built and released Humanity’s Sixth Sense, a benchmark th
I built MaRN (Mapping Networks) , a PyTorch library that lets you optimize a compact latent representation instead of directly training every model parameter. Some results from my current benchmarks: MNIST CNN: Reduced a 537,748-parameter CNN to 4,080 trainable parameters ( 131.8× reduction ), with accuracy decreasing from 99.07% → 98.10% ( −0.97 pp ). A smaller 107,998-parameter CNN was reduced to 1,872 trainable pa
Ai2 at COLM 2026: Open research, from models to agents for science October 9, 2026 Ai2 This week Ai2 was at COLM 2026 in San Francisco. The work we shared reflects our wider mission: building AI research artifacts anyone can study and extend, and developing AI for science tools in collaboration with domain experts. Our evening gathering gave colleagues from across the COLM and wider open-source community another sett
Nature Computational Science, Published online: 09 October 2026; doi:10.1038/s43588-026-01063-0 Researchers pair active learning with geometric deep learning to predict the outcome and site selectivity of chemical reactions from scarce data, guiding experiments that accelerate drug discovery and expand accessible chemical space.
Abstract With the growing use of Artificial Intelligence (AI) in remote sensing of mass movements, open and suitable datasets for model training and validation are increasingly needed. Although Differential Synthetic Aperture Radar Interferometry (DInSAR) is a widely used technique for studying mass movements, wrapped interferograms remain less exploited, and the importance of geomorphological expertise in their inte
image credit- shutterstock China's XtalPi Holdings, an AI and robotics-driven drug and materials discovery company, has debuted Kodexia™, the industry's first closed-loop small interfering RNA (siRNA) discovery platform combining generative AI, first-principles-driven biological modeling, and automated experimentation. Validating the rapid execution capabilities of this new infrastructure, the company simultaneously
Predicting the dynamic polarizability tensor of organic molecules is essential for simulating light-matter interactions in optoelectronic devices, yet conventional quantum-chemical methods such as time-dependent density functional theory (TD-DFT) are computationally expensive and limit large-scale screening. We present an equivariant graph neural network architecture that predicts frequency-dependent, complex-valued
Humans perceive far more in a scene than what is explicitly depicted: a single glance captures past causes and future trajectories; a quick peek determines if a vehicle can fit between two parked cars; a few seconds of video reveals who holds authority in a room; and a fleeting clip highlights subtle abstract patterns like unwritten rules or hidden labels. This capacity reflects a form of humanity's sixth sense: an i
Computational design of polycrystalline microstructures for enhanced mechanical performance requires repeated high-fidelity simulations over stochastic morphologies, rendering conventional crystal plasticity finite element (CPFE) approaches prohibitively expensive for optimization. Here, we develop a deep learning-driven framework for optimizing polycrystalline microstructures under competing quasi-static and dynamic
Black-box optimization (BBO) arises in many scientific and engineering problems where objective evaluations are expensive and limited. Recent large language model (LLM) agents offer a new way to approach BBO by combining task semantics, computation, optimization tools, and feedback-driven decision making, showing great potential due to the integration with mathematically rigorous tools. However, existing agentic BBO
Large Language Models (LLMs) increasingly ship with explicit "thinking modes", yet their counterpart, "no-thinking", has received far less attention. We study LLMs' no-thinking behavior along two axes. a. How to measure no-thinking? Prior work typically defines no-thinking through proxies such as a disabled thinking mode or the absence of long traces. These proxies are unreliable: disabled thinking modes may still em
OpenAI releases 722 AI-generated mathematics papers as researchers weigh what comes next OpenAI has published hundreds of mathematical manuscripts produced by an unreleased AI model, opening a much bigger debate over how AI-generated research should be checked, understood and incorporated into mathematics OpenAI has released a collection of 722 mathematical manuscripts generated by an internal frontier AI model, cove
In a series of interviews, we’re meeting some of the PhD students that were selected to take part in the Doctoral Consortium at the International Joint Conference on Artificial Intelligence and the 29th European Conference on Artificial Intelligence (IJACI-ECAI 2026) . This time, we hear from Mayra Elwes about her work developing machine learning methods for clinical time-series data. Tell us a bit about your PhD – w
Automated building design must comply with seismic and wind codes and satisfy structural mechanics constraints, yet most existing agents produce visually plausible models without verification grounded in mechanical analysis and code compliance. We introduce the Vibe Building task and propose PE-Loop (Physics-Engine-in-the-Loop), an agent in which a deterministic physics engine is the sole source of evaluation signals
This work examines the transfer of a co-evolved communication mechanism between two robotic agents from a discrete two-dimensional (2D) simulator to a three-dimensional simulator with real physics (3D). The study focuses on whether a communication mechanism co-evolved in a 2D environment retains its functional role after transfer to a 3D physics-based simulator. To support this analysis, the effects of the episode ti
Multi-teacher on-policy distillation (MOPD) is used in two settings. In common-domain composition, several teachers score each student rollout from one prompt domain and their signals form a single target; in routed-domain distillation, prompts from different domains are assigned to the corresponding specialist. Both settings usually transfer each teacher's endpoint policy, which mixes what post-training changed with
Latent world models have shown a remarkable ability to predict future states and to plan in the real world. In practice, however, we lack a principled way to estimate how their capabilities scale with model size, data, and compute, an open problem that slows progress in the field. In this work we present RoboJEPA, a world model based on the Joint Embedding Predictive Architecture (JEPA) and trained on a large-scale d
AI Models & Platforms OpenAI Releases 722 Math Manuscripts From an Unreleased AI Model Add Unite.AI to your preferred sources on Google In a research post published October 6, 2026, OpenAI said it is releasing a broad range of new mathematical results produced by an unreleased internal frontier model, publishing 722 manuscripts organized into 372 families in a public GitHub repository alongside Lean proof formalizati
Jennifer Neville is a partner research manager at Microsoft who’s built a career around understanding and advancing AI for real-world use, and much like the human-AI interactions she’s been studying, her early-career path was multiturn: math, then physics; cognitive science, then work; and finally computer science—despite her best efforts to avoid the field. In this conversation with Principal Applied Scientist Chad
Large language models (LLMs) and vision-language models (VLMs) are usually frozen after deployment, so they do not learn from the cases they solve. This is especially concerning in medicine, where new clinical evidence, updated guidelines, and new therapies can change established practice. Fine-tuning can update the model, but it requires access to model weights and additional training. Parameter-free methods avoid t
Large language models (LLMs) have become increasingly capable problem solvers, but being able to solve a problem is not the same as being able to teach it. Existing approaches to training LLMs as teachers rely on demonstrations, preference data, or predefined pedagogical criteria that specify what good teaching looks like. However, these signals are often not grounded in individual student learning outcomes, where ef
Site-specific fertilizer recommendation systems adapt nutrient advice to location, soil properties, crop type, and production targets, but scientific reuse is constrained when recommendation functions remain accessible mainly through interactive interfaces, outputs are not versioned, and trained approximations cannot be independently loaded or benchmarked. This technical report presents the Turba fertilizer machine l
Remote sensing foundation models (RSFMs) are commonly evaluated using aggregate metrics, which can hide systematic performance disparities across ecological regions. We introduce FairRSFM, a biome-aware benchmark for evaluating ecological group robustness in RSFMs. FairRSFM maps georeferenced samples from 14 terrestrial biome classes into six ecologically meaningful macro-groups and evaluates models under a unified f
Modern information systems, including many agentic workflows, use dense retrieval to explore large amounts of unstructured data. However, dense retrieval relies on surface-level semantic similarity, which is insufficient for increasingly complex search applications. Here, we investigate agentic retrieval that combines the reasoning capabilities of Large Language Models (LLMs) with the efficient corpus exploration of
LLM-based agents are increasingly capable of generating complex 3D structures, with the potential to reshape how objects are designed and realized in the physical world. Yet, producing elegant geometry is fundamentally different from producing objects that can be built and perform their intended functions. Existing evaluations largely focus on geometric quality while overlooking physical realizability. We introduce L
Efficient long-video understanding with vision-language models (VLMs) is often framed as selecting informative frames or visual tokens at a fixed native resolution. We show that per-frame resolution can instead be traded for denser temporal coverage, while front-end decoding latency depends on the size of the candidate pool rather than the final token budget. An empirical study across multiple VLMs and long-video ben
On-policy distillation (OPD) has emerged as a widely used paradigm for post-training large language models, reducing the train--test mismatch of conventional distillation by supervising the student on its own generated trajectories. However, existing OPD objectives remain largely token-local and outcome-agnostic, optimizing teacher--student agreement at each prefix despite reasoning quality being determined at the tr
Masked diffusion language models (dLMs) offer a promising parallel alternative to autoregressive models for complex reasoning. However, they face a distinct credit-assignment challenge, since a few commitments during denoising sharply reduce the uncertainty over the remaining masked positions and shape much of the response. Most post-training recipes for dLMs do not use this signal to decide which tokens to train on:
Large language models (LLMs) increasingly serve as general-purpose interfaces to factual knowledge, but their parameters do not automatically reflect information that changes after pretraining. Knowledge editing (KE) provides a targeted alternative to costly retraining by modifying selected knowledge and preserving unrelated knowledge and general capabilities. Conventional KE uses structured factual triples, whereas
Terminal agents are increasingly deployed beyond software engineering in science and other specialized domains. Constructing training environments requires executable reference behavior and a domain-specific verifier that distinguishes semantic correctness from superficially plausible artifacts. Authoring these components for each task requires repeated engineering and limits reuse. We introduce software-in-the-loop
Reasoning traces improve large language models (LLMs), but current models are trained to reason mostly in English. It has been shown that forcing a model to reason in another language degrades accuracy, even when the reasoning language matches the language of the prompt -- but only for a setting where the model reasons over a short prompt. Here, we ask whether the same holds for retrieval-augmented generation (RAG),
Existing streaming vision-language models (VLMs) continuously perceive and reason over visual streams, but their computational pathways remain fixed throughout inference. Consequently, they cannot adapt computation to evolving scene dynamics, where different future events demand different levels and forms of perception. We show that streaming VLMs inherently possess the ability to anticipate the immediate future, and
World action models integrate future visual dynamics with robot action prediction, but their scalability remains limited by the need for action-annotated robot trajectories. Observation-only videos contain rich evidence about interaction dynamics, but existing approaches typically use them either to pretrain visual representations that must later be adapted for control, or to infer latent actions that are subsequentl
Generative models of human-object interaction are bounded by the data that exists: everyday activities involve several objects, but most captured datasets record one at a time, as multi-object capture is combinatorially expensive. Our observation is that interactions are local, so single-object captures already contain the parts of multi-object activities. We compose them: contact-consistent clips of single interacti
Keyword-matching benchmarks can credit small models for tool use they never perform. We document such a false positive in a matched-architecture pair of Spanish security language models and propose a ladder of strict, cheap diagnostics. A 661.6M parameter model (approx. 65% code/technical text; no dedicated SFT) and a 1,109M model (web-heavy multi-phase curriculum; 6B-token tool-SFT) share decoder, tokenizer, and spe
World action models (WAMs) have emerged as a promising paradigm for robotic control by jointly predicting future visual dynamics and actions from an initial observation and instruction. However, existing WAMs struggle with long-horizon prediction, as generating dense video rollouts is highly inefficient. Some recent WAMs address this by predicting a single future frame without generating the full video, but this appr
Computer-use agents need to reliably ground action targets in complex desktop scenes, where multiple applications, overlapping windows, and visually similar controls compete for attention. Existing training data rarely pair such scenes with dense annotations or vary them in a controlled way. We introduce DeskForge, a controllable desktop environment that composes and explores real applications to generate large-scale
What makes great scientists great? Even as AI systems start to make progress on open problems, scientists remain far ahead of them at sensing which prior idea, buried in an ever-growing archive of research, a new problem needs. To study this skill, we draw on researchers who know firsthand which earlier work advanced their completed projects, with papers serving as pointers to the ideas within. Using our automated pi
As LLM agents undertake increasingly complex, long-horizon tasks, verifying their outputs becomes increasingly challenging. We study how verification capability can be strengthened with a fixed base model, without access to reference answers or grading rubrics at test time. Repeated sampling yields multiple rollouts that can contain complementary correct claims, but we need a reliable verification mechanism to determ
On-policy self-distillation has recently emerged as an effective approach for improving language-model reasoning by supervising students with a frozen or EMA version of themselves that receives privileged information. Its application to multimodal large language models (MLLMs), however, remains largely unexplored. Recent approaches use privileged visual information, such as image crops corresponding to a question, to
Video generation models are increasingly being explored as world simulators for embodied planning and learning. To do so effectively, these models must not only generate visually appealing frames, but also predict how environments dynamically evolve when executing goal-directed actions. While evaluating these capabilities is crucial, existing benchmarks focus mainly on single short actions or step-by-step instruction
Abstract Robotic processing of shoe uppers requires accurate segmentation of thin processing boundaries in 3D point clouds. However, acquiring and annotating large collections of real-world scans is costly. We present a multimodal synthetic 3D point cloud dataset of shoe uppers for robotic processing boundary segmentation. The dataset contains 1,067 samples, including 1,000 synthetic point clouds and 67 real-world te
At a glance End-to-end forecasting: A machine learning pipeline uses forecast-time solar-wind information to generate location-specific risk estimates for 66,935 substations in the continental United States. Physics and place: The system combines Auroral Electrojet (AE) and Disturbance Storm Time (Dst) forecasts with local latitude, geology, and ground conductivity. Advance warning: The pipeline detected nearly 80% o
Video generation has rapidly progressed from short, low-quality clips to high-resolution, long-duration sequences with complex spatiotemporal dynamics. Despite strong generative priors learned through large-scale pretraining, pretrained video models often fail to reliably follow human intent, maintain temporal coherence, or satisfy physical and safety constraints. Compared with image and text generation, alignment in
Legal language models require reward signals that capture not only answer correctness but also the multidimensional quality of legal responses. Existing reward methods, however, often rely on coarse-grained holistic judgments, providing limited domain specificity and interpretability. We introduce LexReward, a taxonomy-driven framework for legal reward modeling. LexReward characterizes legal response quality along th
Large language models rely heavily on human text, which often conveys surface answers rather than the spatial and structural logic behind them. Protein folding is a natural testbed, because one solved structure yields thousands of exactly checkable spatial and topological statements. We ask: can learning to fold proteins teach general models reusable reasoning capabilities? To answer this, we build FoldingCorpus, a p
Scientific software presents a demanding test for computer-using agents based on visual language models (VLMs): completing a research workflow requires interpreting specialized interfaces, manipulating scientific objects, and producing verifiable results. We thus introduce OSWorld-Science, a benchmark and evaluation environment that combines scientifically meaningful tasks, artifact-based evaluation, and an efficient
LLM agents generate intermediate reasoning and actions token by token, making extended interactions slow and computationally expensive. Jev-style models offer fast probabilistic predictions over finite fields, but require those fields to be specified in advance. This requirement limits autonomous task solving, where the available actions must be derived from natural language instructions and adapted through interacti
Reproducibility is essential for scientific research, yet prior work shows that LLM outputs vary with hardware and batching. We identify an overlooked factor: the hidden injection of the current date into system prompts, which users cannot control and which changes every day. Across 9 recent LLMs and 6 datasets spanning multiple-choice QA (MCQA), math reasoning, code generation, and machine translation, performance v
This work evaluates the direct transfer of a co-evolved communication protocol from a 2D simulation to a 3D physical environment, without retraining the network weights. Two e-puck-type robots, controlled by a GRU network with residual connection, were evaluated in a food-seeking task with social signaling. The sensory and motor translation layer required three corrections for stable physical operation, including the
Manipulation behaviors vary widely across objects and scenes, but they share a small set of reusable skills, and planning with these skills helps embodied agents generalize to new tasks. Yet an agent can only plan with skills it knows. Recovering skills from observed experience, the inverse of planning, builds this knowledge over time and yields skill data for training future agents. Vision-Language Models (VLMs) des
Coding agents are beginning to move beyond purely digital tasks to tackle physical-world challenges, particularly in robotics. Existing robotics benchmarks, however, primarily focus on the performance of individual artifacts, such as policies or controllers, offering limited coverage of coding agents' broader engineering capabilities. Real-world robotics extends beyond control: agents must build, integrate, diagnose,
Reinforcement learning has helped AI agents solve increasingly difficult tasks, but high rewards do not always reflect the work users intended. In recent incidents and controlled evaluations across the AI industry, agents trained to maximize reward have accessed unauthorized information, attempted to evade monitoring systems, and even breached sandbox protections to attack external systems. As agents become more capa
On-policy learning has been argued to reduce catastrophic forgetting, produce sparser parameter updates, and improve generalisation. However, existing comparisons between supervised fine-tuning and reinforcement learning vary many factors simultaneously, making the contribution of rollout policy difficult to isolate. We study the effect of rollout policy in a controlled strong-to-weak distillation setting, by indepen
Before invoking external tools, an agentic LLM must select among a K-way action space: executing a call, seeking clarification, answering directly, or declining. While internal activation steering can alter these pre-execution decisions, conventional aggregate metrics obscure where altered states land and what collateral damage they inflict. We present SAKIKO, an auditing framework that formalizes representation repa
Mobile manipulation extends robot interaction beyond a fixed kinematic workspace by making the reachable region itself controllable. This flexibility introduces two central challenges: spatially grounded perception under continuous ego-motion and coordinated control of heterogeneous arm and base actions. Existing approaches strengthen geometry through explicit 3D representations or predictive world models, and often
Multidisciplinary tumor boards integrate multimodal clinical observations and longitudinal patient histories through specialist discussions, yet benchmarks rarely capture these real-world trajectories. We introduce OpenTumorBoard, a benchmark with 611 patient cases and 19,157 discussion turns across ten specialist roles, transcribed from 12,534 minutes of publicly available tumor board recordings on YouTube. The benc
What happens when you shrink a language model twice — once by removing some of its weights, once by storing what’s left with fewer bits? You’d guess the damage just adds up. It doesn’t. The problem: does compression just add up? Small language models are increasingly used for on-device apps that need to run in tight memory budgets [Zhou et al. 2025]. To shrink them enough, people often combine two tricks: pruning, wh
Retrieval-based factuality evaluation, where LLM-generated claims are verified against evidence from authoritative medical corpora, has become the dominant paradigm for scalable hallucination detection in high-stakes clinical settings. Despite the urgency of reliable and transparent medical fact verification, most systems measure performance with aggregate metrics like F1, which obscure where and why failures occur.
AI agents are beginning to automate research and development across the AI stack, from improving training efficiency to optimizing inference. A natural next step is to improve the research efficiency of the agents themselves. When an AI research agent's own code is the object of optimization, each accepted rewrite becomes the agent that the next round edits. We refer to this loop as recursive self-improvement. Its si
Tool-calling LLM agents are increasingly deployed in enterprise applications. However, effective evaluation and optimization require high-quality, diverse task datasets that are often difficult to obtain due to privacy and other constraints. Existing synthetic task generation methods often produce generic tasks that ignore an agent's underlying state or database and fail to reflect real-world usage diversity. We prop
Alex Zhavoronkov, co-CEO of the AI drug discovery startup Insilico Medicine, dresses the same way at all his public engagements: a black padded vest with a small white badge that keeps count of the company’s clinical pipeline. When Fortune sat down with him in Hong Kong in July, he made sure to point out the badge’s latest revision: a small circle beside “Phase 3.” The badge is a provocation as much as a point of pri
Large language models can hold knowledge they do not report. A model may sandbag on a capability evaluation, or answer against what it internally knows, and its outputs alone cannot tell whether it is hiding an answer or simply does not have one. We borrow the Concealed Information Test, a forensic method that identifies guilty knowledge by presenting a suspect with the true detail among plausible decoys and measurin
Abstract Accurate meteorological forecasting is fundamental to public safety and economic stability, but it heavily relies on numerical weather prediction (NWP) models, whose complex outputs are often inaccessible to non-experts. Although large language models (LLMs) show potential in translating structured data into natural language to automate forecast reporting, progress has been constrained by a critical scarcity
Nature Computational Science, Published online: 18 September 2026; doi:10.1038/s43588-026-01060-3 Large language models can potentially serve as human proxies across various roles, enabling nuanced simulations of behavior. This Review categorizes their applications, stressing distinct evaluation methods and validity criteria essential for accurate scientific claims and practical implementations.
Skills can improve the performance of Large Language Model (LLM) agents by providing task-specific procedural guidance, while skill optimization further improves their effectiveness through iterative refinement. However, existing skill optimization methods typically represent skills as unstructured natural-language instructions, creating two key challenges: 1) Unstructured skills often lack explicit workflow-level gu
AI Models & Platforms Insilico Medicine Releases Open Longevity AI Toolkit in Cell Study Add Unite.AI to your preferred sources on Google Insilico Medicine on September 17, 2026 announced the publication of a study in Cell introducing an openly released AI toolkit for aging biology: the LongevityBench benchmark, a family of five compact Longevity-LLMs, and the Longevity Claw agentic research platform. Publication in
Artificial intelligence (AI) in radiology aspires to deliver expert-level diagnosis across diverse clinical tasks, yet existing supervised strategies remain limited in scope. We developed RADAR, a generalist vision-language model trained on more than 400,000 contrast-enhanced abdominal computed tomography (CT) examinations and 15 million anatomy-wise image-text pairs, learning directly from clinical reports without m
AI models need to do more than produce correct answers. How they respond matters too: whether they’re helpful, fair, safe, respectful, and responsive to the people using them. For model builders, the challenge is knowing whether those “prosocial” behaviors hold up in practice—and whether evaluations capture how a model behaves when people interact with it in unexpected ways. Northeastern University MS student Soham P
The paper introduces three interconnected resources for the global scientific community: - LongevityBench, the first open benchmark designed to rigorously evaluate AI reasoning across multiple domains of aging biology; - Longevity-LLMs, a family of compact, open-source language models specifically trained on clinical and multi-omics aging data; and - Longevity Claw, an open-source agentic research platform that auton
ScienceIDE: Turning World's Scientific Codebase into Agent Learnable Environments 摘要 Scientific code repositories encode decades of human knowledge in executable models, methods, and tools. Yet fragmented toolchains, implicit domain conventions, and specialized correctness criteria make this knowledge difficult to convert into reliable learning experience—a challenge we call the scientific experience bottleneck. We i
Abstract Molecular descriptors play a crucial role in representing the structural features of molecules for machine learning-based physical property prediction. However, current descriptors either consider only local aspects of molecular structures or fail to effectively learn nonlocal structural features involving long-distance intramolecular interactions. Here, to address this issue, we present a descriptor named T
Large Reasoning Models (LRMs) achieve strong performance on complex tasks but exhibit systematic inefficiency: they often overthink easy problems and underthink hard ones. Existing approaches based on uniform length penalties or rigid routing incur an efficiency tax, trading reduced computation on easy instances for accuracy loss on hard instances. We formulate efficient reasoning as an instance-adaptive computation
Large language models (LLMs) increasingly participate in scientific evaluation, both as automated reviewers and as assistants to human reviewers. As model-generated reviews enter public data and future training corpora, AI peer review can become recursive: later reviewers learn from judgments produced by earlier models. We study one step of this feedback loop in a controlled setting. Starting from Llama 3.1 8B, we fi
JEPA-Anything: Learning Predictive Models across Different Worlds 摘要 World modeling enables intelligence to anticipate consequences, guide interventions, and learn from interaction. Yet predictive models remain domain-specific: can a common learning principle support world modeling across radically different systems? We introduce JEPA-Anything, a domain-agnostic framework based on orthogonal predictive factorization
On-policy self-distillation (OPSD) lets a language model learn from a frozen copy of itself that sees an answer or a worked solution. Giving the teacher this extra information seems to offer the student more to learn, but how much does it add beyond distillation itself? To isolate that contribution, we construct AMPLE-Math, a reusable suite of 5,319 mathematical problems with six reasoning views that share the same a
Understanding why people age — and how to slow that process down — is an obsession that has stretched across millennia, from mythic quests for immortality to modern science’s hunt for drugs that lengthen lifespan. Now, that quest has entered the AI era. A report1 published today in Cell unveils a bespoke artificial-intelligence system that aims to turbocharge longevity research. The system introduces large language m
Spatial intelligence requires more than describing object locations. Under incomplete observation, models must identify and acquire missing evidence, interpret it in a common spatial frame, and act on it. We introduce VA-Bench to evaluate the complete observe-reason-act-revise loop. General-purpose MLLMs learn procedural context from RGB-only demonstrations, actively select camera viewpoints, issue metric Cartesian c
Recent Omni-Modal Generative Models (Omni-Models) have advanced content generation toward unified modeling of text, images, video, and audio. MiniMax-H3 exemplifies this transition by combining multimodal context understanding with joint audio-visual generation in a shared latent framework. Its unified architecture raises a fundamental question: Can multimodal alignment improve the model's world reasoning, and what n
Enabling robots to adapt to unfamiliar environments as readily as humans remains a moonshot goal of embodied AI. No finite collection of demonstrations can cover every task and situation a robot will encounter, making the ability to learn from context at deployment essential for generalization. Such in-context learning (ICL), however, remains largely beyond the reach of existing robotic policies. The broad agentic ca
ScienceBuddy: Recursive-in-Recursive Self-Improvement for Interactive Scientific Agents 摘要 ScienceBuddy is an interactive scientific research workspace that brings continually improving agents into a researcher's everyday work. It supports researchers in carrying out scientific tasks while turning their requests, feedback and execution evidence into tasks and evaluation rubrics for continual learning. At its core is
On September 8, OpenAI said 10,000 of its agents, running for 88 hours, had produced a 166-page proof regarding the Navier-Stokes equations — which describe fluid movement and have remained an open question since 1934. It's one of the seven Millennium Prize Problems established by the Clay Mathematics Institute in 2000, each carrying a $1 million prize; only one has ever been awarded. OpenAI has said it won't be clai
In a new series of interviews, we’re meeting some of the PhD students that were selected to take part in the Doctoral Consortium at the International Joint Conference on Artificial Intelligence and the 29th European Conference on Artificial Intelligence (IJACI-ECAI 2026) . In this first article of the series, we caught up with Yash Saxena to find out how he is trying to make it easier to understand how external knowl
Objectives in scientific machine learning are often prescribed as a sum of several terms, such as the residual, boundary, initial, and data losses of a physics-informed neural network. In the pullback-corrected scalar auxiliary variable (PB--SAV) method, one scalar tracks the shifted objective while the component gradients build a positive semidefinite curvature correction of rank at most the number of components. We
Machine learning, at its core, is about generalization, not memorization. You hand your learning algorithm a pile of training examples and use them to fit a model. But the goal is not to perform well on the training examples — that's easy, you could just memorize the answers. The goal is to perform well on new examples that you have never before seen. If a model does well on the data it was trained on but poorly on f
Large language models (LLMs) can generate biomedical hypotheses, but it remains unclear whether they truly reason from scientific evidence or simply produce convincing-sounding ideas. To study this, we combine three major biological databases: the Kyoto Encyclopedia of Genes and Genomes (KEGG), Rhea, and UniProt, into a unified biochemical knowledge graph and construct a benchmark of 550 paths connecting enzyme sourc
The report says the Chinese company DeepCybo released the physical foundation model PhysBrain 1.5 on September 9, 2026. It scored 72.5 on 28 public benchmarks, ranking first among participating open-source models, with 14 first-place and 10 second-place open-source results. Its score is within 1 point of GPT-6 Astra (73.3) and Gemini 3.6 Flash (73.0), and above Hy-Embodied-VLM-1.0 (66.0) by about 6.5 points and RynnB
Backdoor poisoning attacks add poisoned examples to otherwise-clean finetuning data, pairing a trigger with a target behavior that the model learns to produce when the trigger appears. Existing evaluations typically fix the number of poisoned examples and sample them at random from a candidate pool. We show that this can severely underestimate worst-case vulnerability: across three LLaMA-3-8B backdoor settings, holdi
Permafrost thaw accelerates the initiation and expansion of retrogressive thaw slumps (RTSs), with critical implications for the terrestrial carbon cycle and land-surface energy budgets. However, long-term datasets characterizing spatiotemporal variations in RTS susceptibility across Northern Hemisphere permafrost regions remain lacking. Here, we integrate 55,673 georeferenced RTS occurrence records with machine lear
Background: Histopathological diagnoses in kidney transplant biopsies, as described by the Banff classification, suffer from substantial inter-observer variability and misinterpretation, making them prone to diagnostic errors. Continuous indices based on BanffNET automated lesion scores are proposed to quantify the continuous phenotypic spectrum of kidney transplant biopsies, complementary to pathologist assessment o
Can financial vision-language models (VLMs) turn chart evidence into reliable action recommendations? Existing hallucination evaluations are mostly claim-centric; they assess whether generated statements are supported, but not whether evidence remains traceable through rationale, confidence, and final action. We introduce E2A-Bench, a 969-query benchmark for financial chart reasoning, constructed from 323 HS300 const
Epoch AI says FrontierMath Tier 4 is now "saturated": by its accounting, when attempts across different models and different time periods are accumulated, every Tier 4 problem has been correctly solved by an AI at least once. OpenAI reports a score of 97.6% for GPT-6 Astra, which solved the one problem that had never been cracked before. When Tier 4 launched on July 11, 2025, the top score on the board was only about
Twenty-five Fields Medalists, with Terence Tao and Deng Yu among the first signers, issued an urgent public statement arguing that the way AI companies are pursuing mathematics is now in "severe misalignment" with the goals of the mathematical community. The statement explicitly does not call for stopping AI or deny its achievements. It raises three concerns: AI is solving problems faster than the community can diges
Visual goal and dynamics prediction can provide language-conditioned robot policies with both a target outcome and a representation of action-dependent scene changes. We bring these predictions into action generation and selection through a shared trajectory model. Dynin-Robotics implements this formulation on Dynin-Omni, an omnimodal masked-diffusion backbone, representing language, visual observations, goals, and a
We demonstrate that training LLMs with RL does not improve performance equally across a dataset. RL shows large improvements on easy problems that an LLM is already good at solving, but small improvements on hard problems. We call this the Matthew Effect in RL for LLMs, after the phenomenon of cumulative advantage from economics and network science summarized as "the rich get richer". The naive explanation is that ha
Reinforcement learning with verifiable rewards is typically performed on-policy, keeping training data close to the current policy but limiting learning to trajectories that the policy can discover itself. Off-policy methods such as supervised fine-tuning, on the other hand, can leverage external knowledge beyond the base model's capabilities, but may suffer from large distribution shift. The key challenge is thus to
Artificial intelligence (AI) is increasingly being applied to the design of simple genetic systems, but its ability to generate whole functional genomes has remained untested. Writing in Science, King and colleagues report the generative design of complete bacteriophage genomes. The authors used the genomic language models Evo 1 and Evo 2, which were pretrained on large datasets of DNA sequences, including more than
In this technical report, we propose Pelican-Sim 1.0, a general world model simulator for embodied intelligence that predicts future observations from visual context and robot actions to support downstream learning and decision making. The model incorporates four key design features: (1) Unified action representation: a 28-dimensional action value space covering most mainstream embodiments, keeping one model valid ac
Objects in post-fire environments often undergo irreversible physical transformations that change their geometry, material state, and visual appearance. Detecting and identifying these remnants is critical for locating hazards, reconstructing pre-incident contents, and inventorying losses. Unlike standard image corruptions, these degradations affect the physical structure of the object itself. To study this setting,
Reacting to sudden physical hazards (catching a slipping plate, dodging a falling knife) is both a meaningful test of embodied intelligence and a hard requirement for deploying multimodal large language models (MLLMs) as the decision coreof household robots. Existing evaluations, however, probe intuitive physics passively through question answering over videos, or target deliberate, long-horizon tasks such as navigat
A research idea may be novel, coherent, and scientifically plausible, yet its proposed method may remain insufficiently specified for faithful implementation. We study the codification readiness of implementation-facing research-method specifications, defined by whether they provide sufficient methodological information for a competent implementer or coding agent to construct the intended method without unsupported a
This study examines the ability of large language models (LLMs) to predict the risk of weather-related forced outages in the distribution grid in a zero-shot framework, without labeled training data. The problem is formulated as a binary severity classification task across three forecast horizons (3h, 6h, 12h), using six years of outage records and high-resolution weather data for a utility service area in central Te
Modeling electrochemical interfaces under operational non-equilibrium conditions is vital for energy technologies but remains bottlenecked by the expensive cost of ab initio methods. Current machine learning potentials, largely designed for closed systems under homogeneous electric fields, are limited in open quantum transport applications. To overcome this, we present an E(3)-equivariant graph neural network that em
Context-specific genome-scale metabolic modeling begins with scoring which of the approximately 10,600 human reactions are active in a patient's tumor. Methods in this literature are routinely benchmarked against activity labels obtained by thresholding the same transcriptomic matrix that is supplied to the model as input. We report a self-audit of our own graph attention scorer, MetaGNN, evaluated on TCGA colorectal
Accurate, reliable, large-scale disease predictions are essential to ensure rice production. Existing disease prediction models often face a trade-off between interpretability and predictive capability, necessitating the integration of mechanistic knowledge and data-driven learning within a modelling framework. Accordingly, we propose a physics-informed hybrid gated recurrent unit (PI-HGRU) model for spatiotemporal d
Long-duration gravitational-wave modelling must resolve fast orbital motion together with slow dissipative evolution while preventing small numerical errors from accumulating into secular phase drift. Here we ask whether the finite-time evolution map itself can be learned as an explicit, differentiable, structure-preserving object and then repeatedly composed through a complete inspiral. We construct three neural-flo
The geometric morphology of deposited filaments can significantly influence the structural performance and stability of 3D concrete-printed (3DCP) structures. However, most finite element (FEM)-based approaches for buildability assessment represent printed layers as simplified rectangles, potentially limiting predictive accuracy. This study proposes a geometry-informed modelling framework that integrates the deep-lea
Introducing WeatherNext 3, our most advanced and accurate global weather AI model Every day, the weather influences billions of decisions. Some are as simple as grabbing an umbrella before heading out the door, but others are far more consequential. Wind, rain, and extreme weather events, like heatwaves and droughts, have cascading impacts across agriculture, global supply chains, clean energy production, and nationa
Black-box optimization problems remain challenging because of large, weakly structured, and high-dimensional search spaces. Existing methods often suffer from poor sample efficiency because they rely on direct candidate generation or trial-and-error refinement. A natural way to improve search efficiency is to use world modeling, which can help identify promising optimization directions before costly evaluation. Large
Electron-phonon interactions govern carrier dynamics, transport, and many optical and quantum phenomena in solids, but finite-displacement calculations within the projector augmented-wave (PAW) framework require up to 6N self-consistent supercell calculations for an N-atom system, making them costly for large, low-symmetry, and disordered materials. We introduce MLLocP, a machine learning strategy that learns the sel
Protein language models (PLMs) provide powerful representations of protein sequence, but their utility for proteome-scale binding-site retrieval remains unclear. Here, we present PocketScope, a training-free framework that represents cavity-lining residues using frozen ESM-C 600M embeddings and retrieves related binding sites through exhaustive late-interaction MaxSim, without pooling or approximate nearest-neighbor
Objectives: Dentists prescribe approximately one in ten antibiotics worldwide, yet antimicrobial stewardship (AMS) remains underemphasized in dental education. Large language models (LLMs) may support AMS training, but their proficiency and clinical reasoning in this context remain unclear. We evaluated GPT-4o's accuracy and clinical reasoning on dental antibiotic prescribing questions, stratified by question difficu
Accurate identification of drug-target interactions is consequential for novel drug discovery and development. Deep learning methods for drug-target affinity (DTA) prediction have shown great promise in accelerating drug discovery and reducing development costs. Although graph neural networks have improved drug representation learning for DTA prediction tasks, many models still struggle to effectively and efficiently
Evaluating physical reasoning in video models is difficult because absolute motion measurements depend on frame rate, object scale, and camera calibration, all of which are often ambiguous or unavailable in generated video. We propose a different approach. When two objects in the same scene obey the same physical law, their motions must satisfy predictable relationships, and these relationships hold independent of ca
Brain-computer interfaces translate neural activity into device commands, yet their performance hinges on preprocessing that remains manual, expert-dependent and poorly reproducible. Large language model agents can automate scientific coding, but existing systems lack the modality coverage, raw-data isolation, experience accumulation and domain oversight that neural preprocessing requires. We introduce EasyBCI, a two
Multimodal evaluations cannot say whether a vision-language model misread an image or misreasoned about it, because every existing method for separating the two places a second model in the loop. We introduce the render ceiling, a model-free reference for benchmarks built by rendering known objects: inverting the frozen cameras and re-solving cross-view correspondence recovers exactly the answer the images support. W
AI can already assist with a surprising amount of scientific work: search literature, analyze datasets, write code, and generate and test hypotheses. The harder problem is building systems scientists can steer as a research project unfolds, especially when new evidence changes which hypotheses to pursue, experiments to run, or questions to investigate next. That challenge was the focus of an August 27 event at Ai2 ma
Scientific progress depends not only on finding solutions, but on learning the rules that explain why they work and using that understanding to design better experiments. We introduce science sandboxes, a framework for studying this capability in AI agents through repeated cycles of experimentation, feedback, and hypothesis revision. Science sandboxes invite an agent to query the natural world in different ways, rang
Large Language Model (LLM) agents have shown promise for automating data-science workflows, yet their end-to-end performance depends critically on the agent harness that represents tasks, manages execution state, constrains output artifacts, and provides evaluation feedback. Existing data-science agents often leave this harness implicit, making results difficult to reproduce, compare, and attribute across heterogeneo
Large language model (LLM) agents are increasingly embedded in scientific workflows for literature analysis, drafting, and review. Existing systems advance autonomous discovery and manuscript generation, but do not resolve the governance problem that arises when ideas, methods, results, and claims propagate through AI-assisted workflows without mandatory human approval or artifact-level traceability. This paper propo
Meteorological reanalysis encodes extreme weather through continuous, physically constrained fields, posing a fundamental challenge for vision-language models (VLMs) whose perceptual assumptions are shaped by natural images. Tropical cyclones exemplify this mismatch: critical properties such as intensity extrema, asymmetry, spatial extent, and physical impacts arise from field-level organization rather than object-ce
This work provides an overview of the different strategies that can be used to evaluate the performance of AI models and agents based on large language models (LLMs) for materials synthesis. After providing a brief overview of the key technologies behind the current generation of AI agents based on LLMs, we summarize the different approaches to evaluating these models in the context of materials science and in partic
Large language models (LLMs) achieve strong performance on mathematical reasoning benchmarks, yet the mathematically meaningful skills underlying their reasoning remain underexplored. We introduce \texttt{SHAPE}, a framework that analyzes Chain-of-Thought (CoT) trajectories through two lenses developed in mathematics education: (1) semantic spaces: the model's evolving mathematical interpretations of a problem (e.g.,
Equivariant graph neural networks provide a principled way to model geometric systems, but efficient first-order architectures remain limited in how vector information can be transformed as it moves across a graph. We introduce \textsc{ESNN}, an Equivariant Sheaf Neural Network that enriches this interaction by learning directed, matrix-valued transport between neighboring vector features while preserving exact Eucli
Bioinformatics software and databases are essential components of modern life science research, yet their mentions in the scientific literature are often inconsistent and difficult to systematically identify at scale. The lack of a comprehensive and up-to-date catalog of bioinformatics resources hinders efforts toward automated biomedical knowledge extraction and streamlined data analysis. Here we present SNAIL, a hy
Scientific agents can produce plausible answers while remaining unable to establish whether the computation behind an answer is executable, recoverable, or reproducible. We present BloClaw, an AI4S workstation built around a simple principle: a scientific agent should know what it can do, show how it did it, and state what remains unvalidated. Each capability declares an execution state, input constraints, dependenci
Phage tail-like elements (PTEs) -- tailocins, bacterial type VI secretion systems (T6SS), and extracellular contractile injection systems (eCIS) -- are contractile nanomachines that bacteria use to kill their neighbors and compete within their micro-ecosystems. PTEs help shape microbial community composition. Most PTE detection tools only detect a single PTE class. Moreover, most tailocin detection methods are largel
AI agents for therapeutic reasoning across biological contexts - Michelle M. Li, Carnegie Mellon University - Microsoft Research New England Generative Modeling & Sampling Seminar Therapeutic hypotheses can transfer across diseases but their relevance depends on biological context. The same target, perturbation, or treatment can produce different effects across cell types, disease states, genetic backgrounds, and pat
Functionally graded lattice structures enable lightweight designs with spatially tunable stiffness and density, but their use in multiscale topology optimization is limited by the cost of repeated computational homogenization. This work presents a machine learning-assisted multiscale optimization framework for regular superimposed lattice structures. The unit cell is formed by combining body-centered cubic, face-cent
Heuristic search underlies planning in autonomous systems ranging from warehouse logistics to robotic navigation, yet generic heuristics fail to exploit the structural constraints that govern constrained spatial domains, causing search performance to degrade catastrophically on harder instances. We study this problem through the two-column Flying Block Puzzle, a rigorously NP-complete spatial planning microworld whos
Sparse autoencoders (SAEs), and other interpretability methods could turn AI models in Biology and other fields into engines of scientific discovery by explaining the superhuman capabilities of those models. However, a latent is only useful if we know three things: whether it is coherent, whether it can be described, and whether that description has predictive power. These questions are routinely conflated. We assemb
In many computational science and engineering problems, repeatedly solving fully resolved physics-based models to design for a quantity of interest (QoI) can quickly become intractable, requiring the use of low-fidelity models to predict the same QoI but introduce errors where some features are neglected or are otherwise inaccurately resolved. We use Gaussian Functional Regression (GFR) to learn a correction to a 2D
Graphical user interfaces have emerged as an important environment for evaluating autonomous AI agents on multimodal interactive tasks. Existing benchmarks such as AndroidWorld and MobileWorld provide strong foundations for mobile agent evaluation, but their application coverage and task design do not yet fully capture the diversity and complexity of realistic mobile use. We present GMA, a benchmark for evaluating ge
Machine learning interatomic potentials (MLIPs) have become emerging tools in molecular modeling and computational chemistry. By learning high-dimensional potential energy surfaces from quantum chemical data, MLIPs enable accurate and efficient predictions of structural, thermodynamic, and dynamical properties. However, such models have limitations in predictions of electronic properties and the effects of static ele
Network comparison using optimal transport is a growing area of research in network science. Unlike standard graph metrics, optimal transport computes both network dissimilarity and a transport plan that explains how one graph morphs into another. In this paper, we review how optimal transport compares undirected, unweighted graphs using three primary distances: the Wasserstein, Gromov-Wasserstein, and Bures-Wasserst
Post-training quantization is often treated as a semantically neutral optimization for edge deployment of Large Language Models. When a full-precision source checkpoint is evaluated and quantization is applied downstream without equivalent re-evaluation, this workflow creates a structural validation--deployment gap: because quantization is a many-to-one mapping over parameter space, source-precision certification doe
The widespread adoption of clinical large language models (LLMs) introduces significant risks of automation bias, premature closure, and clinician deskilling. Current interpretability paradigms, including latent space trajectories, Concept Activation Vectors, and Concept Bottleneck Models, suffer from topological stagnation, metric distortion, and epistemic occlusion, frequently masking intermediate diagnostic uncert
As AI begins to build itself, automating alignment research becomes increasingly important to let safety research keep pace. Although measuring the success of alignment research is enormously challenging, researchers (at Anthropic and elsewhere) have developed benchmarks and automated auditing tools, such as Petri, that quantify common alignment failures, like deception, sycophancy, and jailbreaks. In one of our earl
Physics-Informed Neural Networks (PINNs) have recently gained considerable attention as a mesh-free framework for solving partial differential equations. Nevertheless, their performance deteriorates when applied to strongly coupled multiphysics systems, such as Biot's consolidation model, due to severely ill-conditioned optimization landscapes. In this work, we propose a robust PINN-based solver, termed FS-ENGD-PINN,
Scientific computing is undergoing rapid transformation as advances in artificial intelligence, heterogeneous computing, automation, and data-intensive research reshape not only computational tools but also the institutions, workforce models, and collaborative practices that support scientific discovery. This report synthesizes insights from the 2026 Workshop on Next-Generation Ecosystems for Scientific Computing, th
Accurate background intensity estimation is crucial for precise astrometric measurements in astronomical imaging, particularly in complex scenarios such as those encountered in Cassini Imaging Science Subsystem (ISS) observations of Saturn's ring system. Traditional methods, like polynomial fitting and statistical method, often fail in non-uniform conditions, such as those caused by Saturn's rings or scattered light,
Protein language models (pLMs) encode information about protein sequences which enable downstream tasks such as structure prediction, but their internal representations are not well understood. Sparse autoencoders (SAEs) provide a promising tool to disentangle latent pLM representations into interpretable features, but existing annotation pipelines largely rely on protein-level annotations derived from database label
Lyapunov's direct method is a cornerstone of stability and control, but it hinges on finding a Lyapunov function, a task demanding ingenuity or computation. A key difficulty is that every sub-level set of the function must be forward invariant, coupling its geometry to the system's trajectories. We relax this by replacing invariance with recurrence: a set is recurrent if every trajectory starting in it returns within
While Physics-Informed Neural Networks (PINNs) have emerged as a transformative paradigm for solving complex differential equations, their reliance on backpropagation-based gradient descent and automatic differentiation (AD) imposes significant computational bottlenecks and severe non-convex optimization challenges. To overcome these fundamental limitations, we propose the Physics-Informed Stochastic Configuration Ma
The analysis speed is up to 25 times faster Strategic technology to escape under Big Tech Bio-artificial intelligence (AI), such as Google DeepMind's "Alpha Fold 3," which won the 2024 Nobel Prize in Chemistry, has also emerged in Korea. The speed of predicting protein structure and designing new drug candidates is the fastest in the world. KAIST, the organizer of the 'AI-specific foundation model project' promoted b
Recently, there has been a growing adoption of protein language models (PLMs) in biomedical science. Their embeddings provide a rich numerical representation of protein sequences which achieve state-of-the-art performance on several downstream tasks including protein fitness prediction. However, PLM embeddings are not directly interpretable and, thereby, it remains unclear what features they encode. To gain insight i
Synthesis planning for inorganic materials requires predicting both synthesizability and viable routes by linking microscopic thermodynamic stability with macroscopic synthesis methods, precursors, and processing conditions. Here, we develop an interpretable Physics-Informed Retrieval-Augmented Generation Language Model (PIRAG-LM) for end-to-end inorganic crystal synthesis planning. We construct a material-centered S
Artificial intelligence (AI) promises to accelerate biological research by automating computational analyses. Yet the ability of AI agents to carry out computational biology at the scale of complete research studies has not been systematically evaluated. Here we introduce BixBench3, a benchmark that measures the capacity of AI agents to process raw biological data through to scientific results. We designed BixBench3
Large language models (LLMs) hallucinate numerical values when generating high-stakes meteorological text, posing risks for weather communication. We present AFDBench, an AI meteorologist that generates professional Area Forecast Discussions (AFDs) by reasoning through structured AI weather forecast data from Google's WeatherNext 2. We introduce AFDBench, the first benchmark for evaluating generative meteorological r
An accurate and tractable approximation of the single-point mutation-induced change in protein thermodynamic stability, denoted by DDG, is critical for understanding the genotype-phenotype relationship. Several computational methods have been proposed for this problem; however, limited and error-prone training data and the difficult-to-predict magnitude of structural perturbations make this a challenging task. Conseq
Background Large language models are increasingly proposed to post-edit decoded text in communication brain-computer interfaces and augmentative communication. A fluent model can substitute a different intent than attempted (intent drift). Whether meaning survives or confidence flags failure is unmeasured. Methods In-silico benchmark of 20 open-weight models post-editing text (4,252,326 labeled generations) corrupted
August 26, 2026 Ahmed A. Metwally, Staff Research Scientist, and Zechen Li, Student Researcher, Google Research GlucoFM is a lightweight, self-supervised CGM foundation model that models slower glucose trends and short-term deviations in separate streams, producing transferable representations and setting new performance standards across diverse metabolic prediction tasks, including diabetes risk assessment, insulin
Rapid and accurate fault detection in high-voltage transmission networks is essential for grid reliability and equipment protection. Transmission fault datasets are frequently imbalanced, and certain fault types produce electrical signatures that fall within the normal operating envelope, causing single-model classifiers to fail on safety-critical cases. This paper proposes a hybrid two-stage machine learning pipelin
Multilingual language models can solve the same mathematical problem in different languages, but it remains unclear whether they rely on shared features or on language-specific computations that only produce similar outputs. We study this question in five models from four families using the Multilingual Grade School Math (MGSM) dataset, with problems solved in English, German, French, Spanish, Russian, and Chinese, r
Molecular embedding models can serve as foundational infrastructure for computational chemistry and drug discovery, where reusable vector representations support property prediction, virtual screening, and retrieval. Most molecular encoders are specialist models built around a single molecular view, producing unconditional vectors with no language interface for varying the representation. We ask whether multimodal la
Diffusion-based deep generative models have emerged as powerful tools for inverse materials design. Yet many existing approaches overlook essential chemical constraints, such as oxidation-state balance, which can lead to chemically invalid structures. Here we introduce 'crystal generator with valence-constrained design' (CrysVCD), a modular framework that integrates chemical rules directly into the generative process
Future radio telescopes will generate data volumes that are increasingly difficult to analyse using traditional statistical methods, motivating the adoption of machine-learning techniques. In this work, we present YOLO-Chars (YOLO-based Detection and Characterisation of Radio Sources), a two-stage deep-learning framework for the automated detection and characterisation of radio galaxies in survey images. The framewor
A trained reinforcement learning policy does not determine the complete behavior that users encounter: deployment code still schedules, admits, suppresses, or replaces its proposed actions. We contribute \emph{runtime action interference} (RAI), an AI control mechanism that preserves policy parameters while regulating action pacing and filtering configured action patterns after inference. RAI releases a proposed acti
The rapid advancement of large language models (LLMs) has led to increasing interest in their application to scientific domains such as chemistry. However, existing chemistry benchmarks often provide only a narrow view of model capability, focusing on limited task sets while overlooking robustness to variations in problem formulation and chemical representation. As a result, reported performance may overestimate a mo
We present a new Astro-COLIBRI component that converts free-text GCN Circulars into structured, event-linked follow-up records and combines them with structured reports submitted directly by the community. A continuously running Circular listener associates new reports with transient events, applies deterministic pre-analysis, and invokes a schema-constrained large language model extraction step for photometry, conta
Literature reviews are essential to scientific progress, but rigorously evaluating automatically generated reviews remains difficult because many aspects of research utility depend on expert judgment rather than reference-overlap metrics. We introduce LitReview Arena, a battle-style evaluation platform with a structured protocol tailored to literature review quality: domain experts with AI paper-writing experience co
Pretrained genome language models provide reusable representations for DNA sequence analysis, but turning them into reliable downstream predictors remains non-trivial. Their practical performance depends strongly on fine-tuning recipes, and default recipes reported in prior studies may be suboptimal for new tasks or model backbones, making weak downstream results difficult to interpret. These requirements place a sub
High-quality molecular datasets are essential for reliable machine learning in cheminformatics and bioinformatics, yet dataset quality is rarely assessed systematically and its relationship with downstream model performance remains poorly understood. Here, we present MolJam, an open-source framework for quantitative assessment of molecular dataset quality across five dimensions-structural integrity, data quality, exp
Bacteriophage lytic enzymes and depolymerases are relevant to phage biology, antimicrobial development, and protein engineering, but their sequence and annotation data remain dispersed across general databases, specialized resources, genome-centred collections, and prediction-oriented datasets. We present PhageLysData, an evidence-aware and AI-ready resource constructed through reproducible multisource integration, p
Background. Artificial-intelligence design tools now plan genome-scale edits, and agentic systems execute those plans with progressively less human oversight. Biosecurity controls are limited to two points: refusal guardrails at the foundation model and sequence-identity screening at the synthesiser. The design stage between them, where the plan is specified, remains governed by recommendations rather than any deploy
Graph neural networks (GNNs) are routinely employed for short-range forecasting on multivariate time series with a spatial graph structure. Despite the availability of many alternative datasets, method innovations within this domain are predominantly assessed against a rather limited set of benchmark datasets, most notably Chickenpox, PedalMe, WikiMaths, METR-LA, and PEMS-BAY. The evaluation protocols contain baselin
Off-target cleavage is a central safety concern for CRISPR-Cas9 genome editing, particularly in therapeutic applications where unintended double-strand breaks carry clinical risk. We benchmarked five machine learning classifiers: logistic regression on mismatch-count summary features, a random forest and a gradient boosting model on one-hot-encoded sgRNA/candidate-site sequence pairs, a one-dimensional convolutional
According to Latent Space, since 2022, components of the AI pipeline have been shifting from human-made to model-made, each flip starting with a 'patient zero' at a frontier lab. This trend is characterized as simulation: slightly worse (10%), but 100x cheaper and 10,000x faster. Stage 1 (2022): Reward signal went synthetic with InstructGPT's reward model, followed by Constitutional AI and RLAIF, where AI critiques i
Accurate computational prediction of enzyme function, standardized by Enzyme Commission (EC) numbers, is essential for large-scale genome annotation and generative enzyme design. However, it remains unclear whether state-of-the-art predictors learn the intrinsic structural determinants of catalytic activity or merely rely on global sequence similarity to annotated homologues. To address this gap, we introduce EnzymAR
The 35th International Joint Conference on Artificial Intelligence and the 29th European Conference on Artificial Intelligence (IJCAI-ECAI 2026) named three distinguished papers. One paper, titled "Approximate Strategyproofness in Approval-Based Budget Division" by Haris Aziz et al., proves that the Nash product rule (NASH) has an incentive ratio of 2, thereby circumventing the impossibility theorem by Brandl et al.
Summary: Dynamic scientific machine learning (SciML) models that combine mechanistic ordinary differential equations (ODEs) with machine learning (ML) components have applications ranging from learning unknown biological processes to integrating auxiliary data modalities into dynamic modelling. To enable reproducible and efficient SciML training, we introduce PEtab SciML, an interoperable data format for specifying p
This survey proposes viewing agent evolution as dynamic graph transformation, modeling agent state as a dynamic graph where memories, tools, skills, workflows, and inter-agent relations are represented as typed nodes, edges, and subgraphs updated through schema-constrained rewrites. The authors organize existing dynamic-graph-based methods for self-evolving agents into four taxonomies: node/feature evolution, edge/to
Bioinformatics software and databases are essential components of modern life science research, yet their mentions in the scientific literature are often inconsistent and difficult to systematically identify at scale. The lack of a comprehensive and up-to-date catalog of bioinformatics resources hinders efforts toward automated biomedical knowledge extraction and streamlined data analysis. Here we present SNAIL, a hy
Robustness verification of neural networks is increasingly important, due to their use in many critical domains. In certain scenarios, proof sharing has been shown to accelerate incomplete verification techniques by reusing intermediate-layer abstract states, or templates, across queries. However, questions remain as to the robustness of template-based acceleration across varying network architectures, properties, da
Microsoft Research's Skala 1.1 demonstrates the continuous improvement of their deep-learning DFT approach. Trained on 2.5× more data than its predecessor, it delivers substantially higher accuracy across key molecular simulation challenges, including thermochemistry, reaction kinetics, and molecular structure prediction. Skala is now available in CP2K and is being integrated into Psi4, FHI-aims, ORCA, and VASP, brin
A forecast valid at noon and another at 18:00 can tell us where the weather is going, but not everything that happens in between. Storms develop, move and weaken continuously, yet many state-of-the-art artificial intelligence (AI) weather forecasting systems provide forecasts only every six hours. While this resolution is sufficient for many applications, finer temporal detail is essential for applications such as fl
Researchers increasingly use AI to measure variables for downstream analyses, but ignoring prediction errors—even with accuracy above 90%—can cause substantial bias and invalid confidence intervals. Existing remedies like design-based supervised learning and prediction-powered inference rely on costly gold-standard labels. This paper introduces debiased inference with multiple imperfect measurements (DMM), a framewor
Monroe is a new molecular foundation model (MFM) pretrained on over 81 million molecules from the PM6 quantum chemistry dataset. It incorporates innovations including an improved graph representation of stereochemistry, improved training losses (conformer denoising and embedding decorrelation), improved multi-task learning, and uses a prior-data-fitted network (TabPFN) for downstream in-context prediction. Across est
Machine-learned models are replacing first-principles calculations across materials discovery, and physical symmetry is the central guarantee built into them. The debate over how much symmetry to hard-wire rather than learn has run on rotations, where a symmetry error is an approximation error. Some constraints are exact: symmetry forces certain property tensors to exactly zero, so a nonzero prediction is physically
Large language models often fail to answer questions about a bounded document collection without retrieval at inference. This work frames the problem as document knowledge internalization, converting a fixed corpus into usable parametric knowledge. The proposed IAR (Inject, Align, Recover) framework separates structured knowledge injection, QA alignment, and general ability recovery. Inject uses continuation, rewrite
Public benchmarks are important for measuring Automatic Speech Recognition (ASR) capabilities, but models may be optimized for these benchmarks in ways that fail to generalize. This paper presents a methodology to quantify such benchmark optimization, focusing on cases where audio underdetermines the reference transcript. The authors identify three families of behavioral probes—reference disagreement, masked-number r
Self-supervised learning has driven audio representation learning, but existing methods often rely on complex pre-training recipes. The authors introduce NAPE, a framework where a causal Transformer predicts the next patch embedding of a log-mel spectrogram from previous ones, using only causal masking and stop-gradient as training signals, avoiding reconstruction decoders, acoustic tokenizers, and other components.
Existing code generation systems often assume a predefined repository architecture, which fails in zero-to-all code generation where an agent must build a complete software project from natural-language requirements while maintaining modular architecture. We present Repo0, a continuous structural evolution framework that maintains an explicit architectural state as a Dual-Directed-Acyclic-Graph (Dual-DAG), combining
MemTrapBench is a new benchmark for evaluating cognitive traps in large language model (LLM) memory use. Existing memory benchmarks mainly assess whether information is correctly extracted, stored, and retrieved, but overlook how retrieved memories reshape model reasoning and affect current task performance. MemTrapBench covers two forms of cognitive traps: Reasoning Fixation and Belief Distortion. Experiments across
Missense mutations and post-translational modifications (PTMs) are major molecular perturbations that reshape protein function but are traditionally studied independently. Current computational approaches largely rely on sequence conservation or static structural features, limiting our understanding of how perturbations alter intrinsic protein dynamics. We present DynGeo-Pheno, a unified geometric deep learning frame
In this episode of AI Pioneers, Cynthia Rudin, a pioneer in interpretable machine learning and winner of the 2022 Squirrel AI Award, discusses her work on black boxes and Rashomon sets. She recounts her early project with Con Edison on the NYC power grid, where black-box models failed and simple interpretable models outperformed them. Rudin argues that for high-stakes problems like loan decisions or crime prediction,
What happens when you teach an LLM-based agent the scientific method? Motivation: Scientific discovery emerges from cycles of hypothesis, implementation, empirical testing, and feedback. Can this process be automated? We approach automated algorithm design through the lens of the scientific method, where an LLM-based agent goes through each step of the process in an ordered, iterative fashion. Results: We present The
Combining the answers a large language model (LLM) samples for a question into one decision is a test-time information fusion problem, usually solved by majority voting. Voting is unreliable on difficult questions, where the sampled answers share correlated errors, so the wrong answer can win and drawing more samples makes the decision worse. Selecting a candidate by reading a correctness signal from the model's hidd
Physics-informed neural networks (PINNs) offer a flexible framework for forward and inverse problems, but their direct use in structural dynamics is limited by high dimensionality and model-form errors. Existing PINN-ROM couplings often rely on fixed reduced subspaces or assume complete physics. This work proposes a Reduced-Order Physics-Informed Neural Network (RO-PINN) with adaptive basis refinement, embedding redu
Drug response prediction (DRP) models are an active area of research in pharmacogenomics, with growing potential to accelerate the identification of effective anticancer drugs. However, their predictive performance is often constrained by limited dataset scale and insufficient coverages of cancer and chemical spaces. In addition, inconsistent benchmarking practices hinder reliable comparison across models. Standardiz
GEqTrain is a configuration-driven framework that separates dataset semantics, model composition, and training objectives, enabling reuse of equivariant graph neural networks across diverse 3D scientific tasks. Raw data are mapped to typed fields, and models, losses, and workflows are assembled declaratively via Hydra configurations. The framework is demonstrated on three tasks within one software stack: coarse-grain
Reward modeling for joint video-audio generation typically combines metrics for audio quality, visual fidelity, and synchronization, which evaluate perceptual dimensions separately and fail to capture the overall semantic and temporal coherence among text, video, and audio, encouraging reward hacking. To address this, the authors construct a large-scale human-preference dataset VAPref-10K with 9K prompts and 10.3K fi
Perturbational transcriptomics is a powerful tool for understanding gene function and drug effects, yet predicting how perturbations manifest across different biological contexts remains a central challenge, limiting translation from model systems to clinically relevant tissues. Despite growing interest in this problem, benchmarking efforts have been hindered by inconsistent evaluation tasks, heterogeneous metrics, a
Abstract Background: Cardiovascular-kidney-metabolic (CKM) syndrome is an increasingly prevalent multisystem condition associated with morbidity, fragmented care, recurrent hospitalization, and rising healthcare costs. While cardiovascular risk models estimate future disease risk, fewer frameworks support multidisciplinary CKM care, clinician decision-making, and population health management. Synthetic data environme
When a model reads a drug's class from its name—not its knowledge August 18, 2026 Ai2 People increasingly bring health and medication questions to AI models, often naming specific drugs. But a model can sound knowledgeable without knowing much about the drug itself. Many drug names encode their class through affixes – often suffixes like -pril, -olol, and -azoline – giving a model a shortcut: infer the answer from th
We propose a methodology based on the standard ReLU Deep Neural Networks (DNN) to make predictions and quantify their uncertainty. Classically, people rely on linear, non-linear, or non-parametric kernel methods to fit and then predict the time series. As the universal approximation ability was revealed for DNN, its application has become more and more popular for prediction tasks in various scientific areas. However
Bayesian optimization (BO) is a sample-efficient, surrogate-based approach to black-box optimization (BBO), but its evaluation remains dominated by synthetic functions and hyperparameter optimization (HPO) tasks that are typically low-dimensional and single-objective. Engineering design poses a substantially different regime: problems are physics-based, often high-dimensional, constrained by requirements such as cost
Bulk materials, as opposed to nanomaterials, require molecular dynamics (MD) simulations on a large spatial scale (~10^9 atoms or more) to adequately capture their atomic-scale physical properties. Previously, the introduction of machine-learning interatomic potentials (MLIPs) has extended MD to this scale, but even single-component bulk systems require tens of thousands of GPUs on high-end supercomputers. However, m
This work presents a machine learning-symbolic regression (ML-SR) strategy to develop a physically interpretable formula for predicting low pressure CO2 adsorption capacity in hypothetical metal-organic frameworks (hMOFs). Four ML models were trained on a small dataset of 1,000 samples, and five key descriptors—largest cavity diameter, pore limiting diameter, void fraction, gravimetric surface area, and number of hyd
Large language models (LLMs) are increasingly evaluated and used in medicine, but clinical usefulness depends on answer accuracy and whether confidence tracks evidence quality and uncertainty. We developed a controlled, psychophysics-inspired clinical benchmark to test diagnostic choice and confidence behavior in a medical LLM. The benchmark focused on probable Alzheimer-type neurocognitive disorder (AT-NCD) versus d
AI efficiency has recently taken the spotlight in both academy and industry due to massive model scales, high energy demands, and environmental costs. While reporting Floating Point Operations (FLOPs) is a traditional approach for assessing computational costs, the relationship between FLOPs and execution time is not straightforward, as layers with the same number of FLOPs may not have the same execution time because
We introduce PTXBench, a benchmark for evaluating and adapting large language models (LLMs) to use architecture-specific PTX for GPU kernel optimization. PTXBench measures functional correctness, whether selected target instructions execute at runtime, and speedup over frontier libraries across GEMM and attention workloads on H100 and B200 GPUs. Our evaluation shows that architecture-specific PTX capability remains u
This paper introduces Chain-of-Experience (CoE), a paradigm enabling large language models (LLMs) to improve at test time through iterative interaction with self or environmental feedback. Across math, coding, and knowledge domains, using 8 LLMs including GPT-5, Gemini-2.5 Pro, and Claude-4.5 Sonnet, the authors show that leveraging iterative experience consistently outperforms feedback-free baselines, with self-feed
We introduce LifeSciBench, a benchmark of 750 expert-authored tasks designed to evaluate whether language models can handle realistic life science research work. The majority of existing life sciences benchmarks have a narrow scope or are purely knowledge-based, and therefore fail to capture the complexity of real-world research, which often involves ambiguities and requires the accurate execution of multiple depende
We introduce Semantic Task Completion Video Generation, an outcome-oriented task requiring both the achievement of the intended outcome and semantic grounding, which characterizes the correspondence between the reference image and the generated outcome in terms of high-level task-relevant semantics. Evaluation focuses on the generated outcome and does not require a complete sequence of intermediate steps or appearanc
Florent Delgrange won the Best Blue Sky Paper Award at AAMAS 2026 for his work 'Foundation World Models for Agents that Learn, Verify, and Adapt Reliably Beyond Static Environments.' The paper proposes a research agenda that merges reinforcement learning and reactive synthesis into a single loop: a foundation world model is learned from experience but structured for a verifier to reason about. As the agent learns a p
In mixed traffic, decision-making for autonomous vehicles faces three interrelated challenges: physics-based priors fail to capture latent interactive intentions and diverse driver behaviors; abrupt maneuvers cause non-stationarity, leaving long-tail safety events under-explored; and hybrid action spaces destabilize unified RL training due to temporal scale differences between continuous car-following and discrete la
Frontier LLMs are updated frequently and typically outperform their predecessors in aggregate. But aggregate gains say little about individual samples: an update can still cause sample-level regression, where a response correct under the old model becomes incorrect under the new one. This paper studies how to predict such regressions from signals available at inference time. We compare single-model signals (confidenc
Post-training papers, model cards, and blog posts often treat scores on a small set of coding benchmarks (e.g., SWE-bench and LiveCodeBench) as evidence of broad coding capability, both for research artifacts and user-facing systems. We argue that optimization for these benchmarks leads to measuring task-specific performance, creating a meaning gap between measured scores and claims of general coding ability. We exam
Machine learning is rapidly reshaping constitutive modeling, offers new ways to learn material behavior directly from experimental data, and challenges long-established modeling paradigms. But with a growing number of machine-learning-based approaches available, how do they compare in practice? In this paper, we use the classic experimental data of Treloar to benchmark popular frameworks for hyperelasticity: (General
This paper introduces a common layer equation that unifies many graph neural network architectures via seven components: update domain, channel set, propagation bank, per-channel message maps, channel-fusion operator, ego/residual map, and update map. The central factorization separates where information moves (propagation bank) from what moves (message maps). Function-valued fillings extend the equation to local mes
Methods for improving knowledge use in LLMs typically fall into non-parametric retrieval or parametric adaptation. Engram-style hashed memory stores learned information in an external, addressable table, consumed through a small learned reader. This study investigates what matters more when such memory is transferred across backbones: the frozen memory or the target-side reader, via cross-model frozen-memory extracti
PixRestore is a VAE-free pixel-space Diffusion Transformer for unified image restoration (UIR), performing flow matching directly on patchified pixels to preserve fine details. It adapts to degradations by predicting layer feature reliability via LQ-HQ DINO similarity, fusing reliable layers as dense conditioning while supervising less reliable layers to remove degradation. Trained on a large-scale corpus and finetun
Reliable uncertainty quantification (UQ) is essential for deploying large language model (LLM) agents in complex interactive environments. Existing UQ methods often rely on local signals, such as token probabilities or predictive entropy, overlooking long-range dependencies where errors accumulate across an execution trajectory. We propose RUPA, a trajectory-level UQ framework that represents an execution history as
As text-to-image generative models advance, they raise safety concerns, particularly generating Not-Safe-For-Work (NSFW) content, further exacerbated by red-teaming attacks. Existing defenses often rely on white-box assumptions, while black-box alternatives based on LLM prompt rewriting fail in what we identify as the benign adversarial problem: prompts that are linguistically safe but still trigger harmful generatio
Generative artificial intelligence is accelerating molecular design, yet the relative suitability of available models for different targets and stages of preclinical drug discovery remains unclear. Here we benchmarked 12 molecular generation and optimization methods across 176 curated protein ligand systems spanning diverse therapeutic target classes, with experimentally validated ligands providing reference chemical
Looped language models have shown promising results on reasoning benchmarks, yet their potential for agentic tool use remains largely unexplored. We study this question in compositional tool-calling settings, where models must coordinate multiple API calls, maintain intermediate state, and preserve dependencies across tool interactions. We evaluate native and retrofitted looped language models on API-Bank, BFCL, and
Synaptic vesicle glycoprotein 2C (SV2C) is a vesicular protein enriched in dopaminergic neurons of the basal ganglia that modulates dopamine storage and release, and its disruption is implicated in Parkinson's disease (PD). Despite strong genetic and pathological links to PD, there are no selective small-molecule probes for SV2C. Here, we describe an AI-enhanced virtual screening (VS) and experimental campaign that i
Prime Intellect ran 153 internet-isolated autonomous runs across 18 frontier models on the nanoGPT optimizer speedrun, releasing the traces, monitor reports, per-run ledger and harness.
LLM-based agents act on behalf of users to access cloud services, call tools, or invoke agents, but permissions remain static and each request is evaluated independently, allowing deviations from delegated tasks, prohibited combinations of actions, or unrestricted sub-delegation. The Agentic Principal Chain (APC) tracks delegated authority and evaluates requests against accumulated session state using six checks, car
Scientific discovery often involves optimizing expensive-to-evaluate objectives over vast, structured, and open-ended hypothesis spaces, such as molecules, protein sequences, and computer programs. Generative models like large language models (LLMs) provide expressive priors, but their likelihoods and self-assessments are unreliable for objectives and calibrated epistemic uncertainty, especially for novel candidates.
Codon optimization uses synonymous sequence changes to improve the expression and therapeutic performance of nucleic acid-based medicines. Masked language models (MLMs) have recently been proposed as alternatives to traditional, frequency-based codon optimization approaches, yet whether they offer a meaningful advantage over such simpler methods remains unclear. Here we benchmark three prominent MLMs --- CaLM, EnCodo
WorldRover is a data engine for generating richly annotated, long-range explorations of artist-built environments. At its core, WorldRover-Engine is an Unreal Engine pipeline that executes and offline-renders minute-scale routes while preserving their full trajectories and scene geometry. The same exploration can be replayed from first-person, third-person, and 360 panoramic cameras under different environmental stat
Multimodal large language models increasingly use visual chain-of-thought (Visual CoT), which generates intermediate reasoning images for spatial, temporal, and embodied reasoning, but introduces significant inference overhead. This paper introduces Internalized Visual Thinking (IVT), a post-training framework that jointly optimizes textual prediction and next-embedding prediction on unlabeled videos, enabling the mo
Memory is becoming core infrastructure for long-horizon LLM agents, yet existing evaluations offer limited guidance on which memory substrate—the underlying medium in which memory is represented and stored—should be used under different operating regimes. We present a controlled harness evaluation of memory substrates for memory-augmented agents, covering dense and sparse indices, text records, structural stores, hie
Audio-Text Foundation Models (ATMs) are prone to catastrophic failure under severe acoustic noise, but existing adaptation methods either reinforce noise instead of signal or require privileged noise annotations. PRISM (Prototype-Rectified Iterative Self-Supervised Manifold Denoising) is a training-free, source-free TTA framework that leverages the Affine Noise Hypothesis, positing that severe acoustic noise induces
The rapid advancement of artificial intelligence has substantially increased the demand for large-scale datasets; however, available datasets remain insufficient for data-intensive tasks, particularly in nuclear medicine. While large-scale datasets exist for computed tomography (CT), few provide paired projection-domain and image-domain data across diverse lesion types. This issue is more pronounced for positron emis
MOSS-VL is an open vision-language model family that treats real-time interaction—perceiving while speaking—as a first-class capability. It is co-designed across the stack: the language decoder attends to vision only through gated cross-attention, enabling natural perception of incoming frames during generation; a synthesized interaction corpus supervises when to speak, stay silent, or revise; and a staged curriculum
Autoformalization is commonly framed as translating natural-language mathematical statements into machine-verifiable formal languages such as Lean 4. However, faithful formalization requires more than translation; models must map concepts to the complex hierarchy of types and definitions in formal libraries like Mathlib. Existing approaches struggle because they rely heavily on parametric memory and lack feedback-dri
DeepMind’s Pushmeet Kohli on AI’s Scientific Revolution Pushmeet Kohli leads AI for Science at DeepMind, where his team has created AlphaEvolve, an AI system that discovers entirely new algorithms and proves mathematical results that have eluded researchers for decades. From improving 50-year-old matrix multiplication algorithms to generating interpretable code for complex problems like data center scheduling, AlphaE
Open-weight language models are often fine-tuned, quantized, pruned, or merged without documented provenance. We propose a data-free, white-box method for lineage verification: by removing the identity-aligned component that residual training imparts to branch products, we isolate checkpoint-specific structure across residual blocks and compute a symmetric lineage score. On residual-MLP and GPT-2 benchmarks, the scor
AI co-scientists are beginning to transform research by generating hypotheses, retrieving related work, designing experiments, executing code, and drafting papers, but current systems ignore the individual researcher, optimizing generic metrics like novelty and reviewer scores. This paper introduces personalized auto-research, which conditions every stage on a representation of the researcher, arguing that personaliz
High-quality creative writing data for LLMs is dominated by story-centric data, limiting models' adherence to diverse creative formats. The authors propose an attribute-guided genre expansion framework that uses human-authored story prompts as creative seeds and manually curated genre attributes to enforce structural, stylistic, and formatting conventions. This produces genre-faithful query-response pairs, which are
Popular facts are memorized more deeply during pretraining and resist removal longer than rare ones, yet existing LLM unlearning methods apply uniform gradient pressure regardless of training-data frequency. We propose AdaPop (Adaptive Popularity), which combines local token confidence with a per-fact popularity-dependent exponent derived from an external proxy (e.g., Wikidata sitelinks, LLM-as-Judge), and automates
AutoResearchEval introduces 100 tasks spanning 7 scientific domains and the full research lifecycle, including ideation, retrieval, execution, analysis, writing, and review. Evaluating 8 harness-model combinations yields 800 trajectories with process-level annotation, leading to the AutoResearch Failure Taxonomy (ARFT), a framework of 45 empirically-grounded failure patterns. A human-calibrated agent-as-a-judge pipel
LLM coding agents issue Bash commands through interfaces that may serialize, wrap, and reparse model output. Matched execution scores alone cannot distinguish command-generation errors from failures introduced after generation. QuoteBench measures this boundary with exact final-state validation on 56 one-shot tasks from 14 incident-derived families, crossing the generation contract with the execution transport around
Existing AI scientist systems typically reason over text, code, or precomputed summaries, missing crucial spatial, temporal, and cross-channel evidence. To address this, researchers introduce OmniScientist, an end-to-end omni-modal AI scientist that directly conducts multidisciplinary research from heterogeneous raw evidence. The system features a perception layer and three autonomous agents for ideation, experiment,
CardioState-JEPA is a cardiac foundation model that learns a shared representation across ECG, PPG, and PCG using a physiology-aware joint-embedding predictive architecture. It maps heterogeneous waveforms into a common token space, processes them with a shared Transformer encoder, and learns by predicting masked latent cardiac states. A learned delay aligner handles temporal offsets between electrical, mechanical, a
At a glance MindTopo is a new benchmark for testing topological reasoning in AI, evaluating whether multimodal models can understand concepts such as connectivity, enclosure, order, separation, and knots. The benchmark measures both reasoning and planning, testing not only whether models can recognize topological relationships in static images but also whether they can preserve and manipulate those relationships thro
AINews reports on a new paper demonstrating how encrypted reasoning traces from frontier models can be decoded and replayed to other models, sessions, or users, leading to dramatic improvements in open models. Since the release of o1, reasoning models have obscured their traces with cryptographic signatures to prevent distillation, but this study shows that these encrypted blocks can be replayed to weaker models and
Transition states (TSs) govern the rates and outcomes of chemical reactions, making their accurate prediction a central challenge in computational chemistry. Although recent machine learning models achieve near-chemical accuracy in the prediction of TS structures and the associated reaction barriers for small organic reactions, their ability to generalize beyond the training domain remains largely unexplored. Here we
Creative agents still lack an effective way to learn from high-quality human films, limiting their ability to produce cinematic-grade videos. To address this, researchers propose the Agentic Video Auto-Encoder (AVA-Encoder), a framework for learning agent-native video representations via agentic auto-encoding. AVA-Encoder transforms a video into a knowledge graph (KG) representation and reconstructs it back into vide
Multiple-choice benchmarks are widely used to evaluate large language models, but MCQ scores conflate knowledge with sensitivity to option order, making them unreliable measures of model knowledge. This paper tests whether preventing the model from seeing option labels while committing to an answer removes positional influence and improves performance. Two strategies are evaluated: generation-then-matching and scorin
Microsoft Research introduces CARE-X, a unified chest X-ray vision-language model (VLM) designed for diverse clinical interpretation tasks. It combines free-text generation with structured prediction to provide both narrative reports and deterministic outputs. The model uses reinforcement learning (DAPO) to reward clinical correctness in a multi-task setting. In a separate research experiment, the team paired Qwen3-V
In this crosspost from AI Matters, a publication of ACM SIGAI, Ella Scallan interviews Akari Asai, winner of the 2025 AAAI Doctoral Dissertation Award. Asai's doctoral research focused on retrieval-augmented language models (LMs), arguing for moving beyond monolithic LMs to augmented LMs that are designed, trained, and deployed with complementary modules. She pioneered retrieval-augmented LMs, which retrieve relevant
An important article by field experts points out that despite widespread AI application in drug discovery, evidence of clinically relevant impact remains 'disappointingly limited'. They conclude we are at a stage of 'absence of evidence', not necessarily 'evidence of absence'. Desired outcomes like better decision-making in selecting disease areas, targets, lead molecules, and trial designs are among areas current AI
Today we're introducing a preview of TutorMoments, a framework to measure whether cutting-edge LLMs can balance one of the hardest trade-offs in education: when to step in and help a student and when to hold back and let the student do more of the work. TutorMoments is a replay-based evaluation built off real one-on-one math tutoring sessions. Experienced math teachers go through transcripts collected from a U.S. tut
Ai2 expands collaboration with Hugging Face to accelerate open science August 6, 2026 Ai2 Ai2 and Hugging Face are expanding their partnership to make fully open AI easier to access and deploy. Hugging Face serves as a central platform for the AI community, and the expanded collaboration gives us substantially more storage and bandwidth to support our growing portfolio of models, datasets, benchmarks, and application
CogTwin, presented at IJCAI 2025, is a hybrid cognitive architecture framework designed to bring autonomous reasoning and real-time adaptation to digital twins. It addresses the gap in current digital twins that are largely reactive, relying on pre-programmed rules and static models, and thus struggle with unforeseen events. By integrating symbolic AI, sub-symbolic AI, and neuro-symbolic AI, CogTwin aims to equip dig
Artificial intelligence has significantly expanded candidate generation in drug discovery, but selecting optimal compounds under real-world experimental constraints remains a major bottleneck. This review examines quantum annealing as a potential downstream optimization layer within hybrid computational workflows. By focusing on constraint-dominated decision tasks formulated as quadratic unconstrained binary optimiza
In healthcare settings, LLM performance differs between evaluation and deployment, with a recent study finding a 61 percentage point gap. The authors argue this gap arises not from poorly designed benchmarks but from implicit assumptions in evaluation protocols that fail at deployment. They propose a taxonomy categorizing assumptions into task and outcome types to diagnose where the gap arises and what is needed to c
Microsoft introduces Echoverse, a suite of twelve training worlds for computer-use agents: ten deep domain worlds and two capability worlds, each drilling a single control (e.g., date pickers, nested filters) in varied forms. Depth is key: these worlds replicate real application behavior, come seeded with realistic data, and maintain coherent state across screens and users. Training on all twelve nearly doubles a 9B
Microsoft Research introduces EvoLib, a framework that enables large language models to learn from their own experience during inference without ground-truth labels or external feedback. EvoLib transforms past attempts into reusable skills and reflective insights, continually refining, consolidating, and reweighting them to turn instance-specific observations into increasingly general knowledge over time. This evolvi
A study led by the Australian National University (ANU) Emotions and Faces Lab has shown that humans can be trained to improve their ability to spot AI-generated faces. Deepfake faces have become so realistic that they are hard to distinguish from real photos, contributing to AI-related fraud. The training, which focuses on six perceptual qualities—distinctiveness, memorability, proportionality, symmetry, attractiven
The OlmoEarth Platform: Geospatial inference at planetary scale July 28, 2026 Ai2 The OlmoEarth models are our family of Earth observation foundation models, pretrained on roughly 10 terabytes of multimodal satellite data. Governments, NGOs, and other mission-driven organizations are already adapting OlmoEarth for applications including deforestation monitoring, food security, and wildfire risk. At Ai2, we know how t
The author recounts attending a workshop on mathematics and AI, which produced the Leiden Declaration, now signed by over 3000 people. Although he did not sign, he clarifies it is not due to disagreement but rather uncertainty about some confident assertions and recommendations. He appreciates the declaration's role in focusing attention on pressing issues, and quotes its passage on characteristic values of mathemati
As AI becomes critical infrastructure for science, medicine, industry, and government, one question matters more than ever: who gets to understand how these systems work? At Ai2, that question is central to our work, and it’s what motivates our commitment to open science and open models. For us, openness is more than a licensing choice—it shapes how we build, release, and evaluate our work. Open-weight models let peo
The notorious Jacobian conjecture can be formulated concretely over the complex numbers as follows. Conjecture 1 (Jacobian Conjecture) Let be a polynomial map in complex variables, whose Jacobian is a non-zero constant. Then is invertible (with polynomial inverse). The condition that the Jacobian is non-zero is equivalent to being locally invertible. (The implication of local invertibility from non-vanishing Jacobian
Traditional robot learning relies on collecting demonstrations and evaluating policies on real robots, which is slow, expensive, and hard to reproduce. We propose an Interactive World Simulator: a learned, action-conditioned video prediction model that predicts subsequent pixel frames given the current image and robot actions, without a physics engine. On a single RTX 4090, it runs for over 10 minutes at 15 FPS, prod
I believe that the creation of visualization apps to illustrate mathematical or scientific concepts is a particularly favorable use case for modern coding agents, as many of the downside risks attached to other LLM use cases are limited: Not mission-critical . As such apps are not authorative sources of truth and only used for secondary purposes, a small positive error rate in the output can be acceptable. Stand-alon
The work audits AI protein structures at the level of electrostatic interactions, beyond aggregate structure scores.
Enzymes drive cellular metabolism, yet predicting catalytic properties from amino acid sequences remains challenging. Existing protein language models (PLMs) provide powerful general-purpose representations but are often inefficient for high-throughput screening and insufficiently adapted to enzyme-specific tasks. Here, we propose EnzGFM, an enzyme-specific PLM based on a Mamba-Transformer hybrid architecture with hi
Our approach to bioresilience The global biosecurity landscape is rapidly evolving. Shifting natural ecosystems, global travel and the potential misuse of AI require greater vigilance — yet AI is also a critical tool for our response. We need frontier AI models, and the scientific advances they will enable, to respond to these challenges and help make society more resilient to events like future outbreaks. Today, Goo
Therapeutic peptides combine the advantages of small molecules and antibodies, offering target flexibility and low immunogenicity, yet their successful translation requires careful evaluation of multiple developability properties beyond binding alone. As chemically modified peptides become increasingly common in drug design, no unified platform currently supports systematic property assessment across both canonical s
Terry Tao demonstrates how coding agents facilitate converting code-generated figures into interactive applets, using Figure 1 from his recent paper on the Gilbreath conjecture with Chase and Hunter. The figure plots exact and Monte Carlo simulated values of a sequence called the "Gilbreath expectation sequence," which arises from a Gilbreath array: an inverted pyramid with top entries as independent exponential rand
Across three datasets, compliant AI analyses still produced widely dispersed conclusions sensitive to model and persona choices.
Universal machine learning force fields (UMLFFs) promise to revolutionize materials science by enabling rapid atomistic simulations across the periodic table. However, their evaluation has been limited to computational benchmarks that may not reflect real-world performance. Here we introduce UniFFBench, a comprehensive evaluation framework featuring the MinX dataset-a diverse collection of 1,500+ mineral systems span
Protein language models, as well as models that incorporate structural information or homologous sequences, estimate the sequence likelihood p(sequence), which reflects the protein fitness landscape and is commonly used in mutation effect prediction and protein design. It is widely believed in deep learning field that larger models perform better across tasks. However, for fitness prediction, language model performan
This week Tony Wyss-Coray and I published an invited review paper in Nature Medicine entitled “Biological Aging Clocks in Health and Disease.” The concepts and content of the paper may be difficult to understand for people not working in life science, but they are groundbreaking and have exciting implications for the future of medicine. So this edition of Ground Truths is intended to make all of the progress comprehe
This paper examines task construction as a control surface in continued pretraining. Standard pretraining fixes the self-supervised objective before training, with downstream feedback only available after the entire run, making the control loop coarse. The authors propose value-based pretraining (V-pretraining), which separates the learner from a lightweight task designer. The learner updates only via self-supervised
In the latest interview with AAAI/SIGAI Doctoral Consortium participants, Thi Kieu Khanh Ho, a PhD student at McGill University and Mila – Québec AI Institute, discusses her research on time-series anomaly detection. Her work focuses on teaching AI systems to detect unusual events in complex real-world data streams without relying on large amounts of labeled examples, which are often scarce, noisy, or contaminated. S
Microsoft Research AI for Science introduces Flint, a visualization intermediate language for AI-driven chart creation. Flint enables AI agents to reliably generate expressive, visually polished charts from simple, human-editable specifications. It leverages semantic data types to guide the compiler in choosing appropriate scales, baselines, formatting, and color schemes, and automatically manages layout to adapt to
NEP89 unifies inorganic and organic systems while targeting simulation speeds orders of magnitude beyond first-principles methods.
Large language models (LLMs) are increasingly used as synthetic survey respondents, but their individual answers may look plausible while failing to preserve the joint distribution, latent structure, reliability, mediation pathways, and demographic effects of real human data. This study introduces a Lithuanian organizational-psychology dataset (n=263 employees; 68 items, 12 subscales) and evaluates 37 LLMs from OpenA
SkillOpt treats an agent's skill file as a trainable parameter outside a frozen target model, converting skill writing from one-shot prompting into a controlled optimization process. Across six benchmarks, seven target models, and three execution modes, SkillOpt achieves best or tied-best performance in all 52 evaluation cells, improving performance without updating model weights. The method keeps skills compact and
Machines have become remarkably good at perceiving motion. Given a video, modern models can track how objects and points move through a scene with exceptionally high confidence. But perception is inherently retrospective: it explains motion that has already happened. Many of the systems and applications we want to build need tolook forward instead. A robot reaching for a cup has to anticipate how the cup will move be
LifeSciBench evaluates models on research tasks designed with life scientists rather than relying on generic knowledge questions.
HarmProfile is a content-centric benchmark dataset that collects model misbehavior across diverse harm categories and model families, defining the resulting harmful-output distribution as a model-level risk profile. It contains over 80,000 validated artifacts from 23 frontier LLMs across 13 model families, organized into 15 harm categories and 57 subcategories. Using this corpus, the authors find that frontier LLMs r
A Stanford-led research team that includes Fei-Fei Li published a benchmark on May 18, 2026, that documents a specific, measurable failure in every frontier multimodal AI model tested: when forced to actively move through a 3D environment rather than interpret a pre-composed image, state-of-the-art systems consistently choose the wrong actions, skip viewpoints that would correct their mistakes, and commit to wrong an
Image credit: Quang / Adobe Stock Individual ID in Chatty Species In the wild, zebra finches don’t sing in isolation. They move through dynamic social groups, calling and responding to one another across space and time. For ecologists, understanding this communication requires knowing who is communicating with whom , and how information flows through a group. Tracking communication at the level of individuals, howeve
Science Apr 29, 2026 In this post, Brianna, a researcher on the discovery team, shares results from a recent bioinformatics benchmarking effort. Almost as soon as large language models could hold a conversation, people started asking how they’d stack up against human experts. Could models pass the bar exam? Could they answer medical licensing questions, or solve Olympiad math problems? Such benchmarks—self-contained
Explaining deep neural network predictions on genome sequences enables biological insight and hypothesis generation-often of greater interest than predictive performance alone. While explanations of convolutional neural networks (CNNs) have been shown to capture relevant patterns in genome sequences, it is unclear whether this transfers to more expressive Transformer-based genome language models (gLMs). To answer thi
The Research and Applied AI Summit (RAAIS) is a community for entrepreneurs and researchers who accelerate the science and applications of AI technology. In the run up to our 10th annual event on June 12th 2026 in London, we’re running a series of speaker profiles to shed more light on what you can expect to learn on the day! Vivek Natarajan is a Research Lead at Google DeepMind leading research at the intersection o
Despite the recent release of numerous protein language models, relatively little work has addressed how to best sample from them to optimize desired biological properties. This paper fills that gap by proposing a flexible and effective sampling method for masked language models (MLMs), and by systematically evaluating models and methods both in silico and in vitro in the context of antibody therapeutics campaigns. T
OpenAI’s GPT can summarize research papers and make predictions—but can it do science? Can it generate hypotheses, design experiments, interpret results and iterate? Last summer researchers at OpenAI and Ginkgo Bioworks, a company that designs and installs autonomous, robot-run labs, decided to find out. Though artificial intelligence systems have posted high scores in math, physics and computer science, biology is h
"AI Has Overwhelmed Human Exam Questions," A Conversation Between Terence Tao and Mark Chen: Large Language Models Are No Longer "Math Underachievers" Recently, at a seminar hosted by the Institute for Pure and Applied Mathematics (IPAM), Fields Medalist Terence Tao and Mark Chen, the Chief Researcher at OpenAI, participated in a "fireside chat." The two discussed the leap - forward progress of AI in the past year an
Optimizing molecules to achieve desired properties is a central bottleneck across the chemical sciences, particularly in the pharmaceutical industry, where it underlies the discovery of new drugs. Since molecular property evaluation often relies on costly and rate-limited oracles, such as experimental assays, molecular optimization must be highly sample-efficient. To address this, we introduce SEISMO, an LLM agent fo
Accelerating Mathematical and Scientific Discovery with Gemini Deep Think Under direction from expert mathematicians and scientists, Gemini Deep Think is solving professional research problems across mathematics, physics, and computer science In the summer of 2025, an advanced version of Gemini Deep Think achieved Gold-medal standard at the International Mathematics Olympiad (IMO) and later, an updated version, obtai
Epoch AI released a pilot of 14 research-level open problems with programmatically checkable answers to test the frontier beyond existing benchmarks.
Share New publication demonstrates that insitro’s POSH platform — integrating pooled CRISPR screening and self-supervised deep learning — successfully breaks the historic compromise between scale and depth to more rapidly identify novel therapeutic targets. SOUTH SAN FRANCISCO, Calif. — Dec. 16, 2025 — insitro, the AI therapeutics company built on causal biology, today announced the publication of research in Nature
Music editing is vital in modern production, yet many systems overlook preserving musical facets that should remain unchanged. The authors define this as Music Context Preservation (MuseCP) and introduce MuseCPEval, the first evaluation framework covering four categories of music facets with fine-grained metrics. Objective validation and human studies confirm effectiveness, while case studies demonstrate utility as a
Today we are releasing RFdiffusion3 as open-source software. This state-of-the-art AI model for biodesign is capable of generating new proteins that interact with any type of molecule commonly found inside living cells. “We built this as a general model and are sharing it freely so the scientific community can create things we haven’t even imagined yet.” Lead developer Rohith Krishna, PhD RFdiffusion3 tackles researc
Molecular Hessians, the second derivatives of the potential energy, are fundamental to many workflows in computational chemistry. Usually, accurate Hessians are computationally expensive to calculate and scale poorly with system size, whether computed using quantum chemistry methods or machine-learning interatomic potentials (MLIPs). In this work, we introduce Hessian interatomic potentials (HIPs), a deep learning mo
FutureHouse's head of science explains its agent stack, a drug-repurposing case, and why scientific publishing and evaluation must be redesigned for machine readers.