Reduced-precision data formats are crucial for cost-effective serving of large language models (LLMs). While numerous reduced-precision formats have been introduced thus far, they often require intrusive modifications to the software frameworks or are rather unconventional for widespread adoption across hardware vendors. In this paper, we instead focus on recent industry-driven variants of block floating-point (BFP) formats and conduct a comprehensive analysis to push their limits for efficient LLM serving. Our analysis shows that existing ultra low-bit BFP variants struggle to provide reasonable language model performance due to outlier values in blocks. To address the outliers with BFPs, we propose MX+, a cost-effective and non-intrusive extension designed for seamless integration into the microscaling (MX) formats. MX+ builds on the key insight that the outlier does not need to use its exponent field in the element data type, which allows us to repurpose the exponent field as an extended mantissa to increase the precision of the outlier element. Our evaluation shows that MX+ achieves significantly higher model performance compared to the 4-bit MX format (MXFP4) with negligible storage overhead and slowdown, thus offering a compelling alternative to MXFP4 or MXFP6 for efficient LLM inference.

Recently, Agentic Reinforcement Learning (Agentic RL) has made significant progress in incentivizing the multi-turn, long-horizon tool-use capabilities of web agents. While mainstream agentic RL algorithms autonomously explore high-uncertainty tool-call steps under the guidance of entropy, excessive reliance on entropy signals can impose further constraints, leading to the training collapse. In this paper, we delve into the challenges caused by entropy and propose the Agentic Entropy-Balanced Policy Optimization (AEPO), an agentic RL algorithm designed to balance entropy in both the rollout and policy update phases. AEPO comprises two core components: (1) a dynamic entropy-balanced rollout mechanism that adaptively allocate global and branch sampling budget through entropy pre-monitoring, while imposing a branch penalty on consecutive high-entropy tool-call steps to prevent over-branching issues; and (2) Entropy-Balanced Policy Optimization that inserts a stop-gradient operation into the high-entropy clipping term to preserve and properly rescale gradients on high-entropy tokens, while incorporating entropy-aware advantage estimation to prioritize learning on high-uncertainty tokens. Results across 14 challenging datasets show that AEPO consistently outperforms 7 mainstream RL algorithms. With just 1K RL samples, Qwen3-14B with AEPO achieves impressive results: 47.6% on GAIA, 11.2% on Humanity's Last Exam, and 43.0% on WebWalker for Pass@1; 65.0% on GAIA, 26.0% on Humanity's Last Exam, and 70.0% on WebWalker for Pass@5. Further analysis reveals that AEPO improves rollout sampling diversity while maintaining stable policy entropy, facilitating scalable web agent training.

Neuro-symbolic (NeSy) AI aims to develop deep neural networks whose predictions comply with prior knowledge encoding, e.g. safety or structural constraints. As such, it represents one of the most promising avenues for reliable and trustworthy AI. The core idea behind NeSy AI is to combine neural and symbolic steps: neural networks are typically responsible for mapping low-level inputs into high-level symbolic concepts, while symbolic reasoning infers predictions compatible with the extracted concepts and the prior knowledge. Despite their promise, it was recently shown that - whenever the concepts are not supervised directly - NeSy models can be affected by Reasoning Shortcuts (RSs). That is, they can achieve high label accuracy by grounding the concepts incorrectly. RSs can compromise the interpretability of the model's explanations, performance in out-of-distribution scenarios, and therefore reliability. At the same time, RSs are difficult to detect and prevent unless concept supervision is available, which is typically not the case. However, the literature on RSs is scattered, making it difficult for researchers and practitioners to understand and tackle this challenging problem. This overview addresses this issue by providing a gentle introduction to RSs, discussing their causes and consequences in intuitive terms. It also reviews and elucidates existing theoretical characterizations of this phenomenon. Finally, it details methods for dealing with RSs, including mitigation and awareness strategies, and maps their benefits and limitations. By reformulating advanced material in a digestible form, this overview aims to provide a unifying perspective on RSs to lower the bar to entry for tackling them. Ultimately, we hope this overview contributes to the development of reliable NeSy and trustworthy AI models.

In text-to-image generation, different initial noises induce distinct denoising paths with a pretrained Stable Diffusion (SD) model. While this pattern could output diverse images, some of them may fail to align well with the prompt. Existing methods alleviate this issue either by altering the denoising dynamics or by drawing multiple noises and conducting post-selection. In this paper, we attribute the misalignment to a training-inference mismatch: during training, prompt-conditioned noises lie in a prompt-specific subset of the latent space, whereas at inference the noise is drawn from a prompt-agnostic Gaussian prior . T o close this gap, we propose a noise projector that applies text-conditioned refinement to the initial noise before denoising. Conditioned on the prompt embedding, it maps the noise to a prompt-aware counterpart that better matches the distribution observed during SD training, without modifying the SD model. Our framework consists of these steps: we first sample some noises and obtain token-level feedback for their corresponding images from a vision-language model (VLM), then distill these signals into a reward model, and finally optimize the noise projector via a quasi-direct preference optimization. Our design has two benefits: (i) it requires no reference images or handcrafted priors, and (ii) it incurs small inference cost, replacing multi-sample selection with a single forward pass.Extensive experiments further show that our prompt-aware noise projection improves text-image alignment across diverse prompts.

Defective surgical instruments pose serious risks to sterility, mechanical integrity, and patient safety, increasing the likelihood of surgical complications. However, quality control in surgical instrument manufacturing often relies on manual inspection, which is prone to human error and inconsistency. This study introduces SurgScan, an AI-powered defect detection framework for surgical instruments. Using YOLOv8, SurgScan classifies defects in real-time, ensuring high accuracy and industrial scalability. The model is trained on a high-resolution dataset of 102,876 images, covering 11 instrument types and five major defect categories. Extensive evaluation against state-of-the-art CNN architectures confirms that SurgScan achieves the highest accuracy (99.3%) with real-time inference speeds of 4.2-5.8 ms per image, making it suitable for industrial deployment. Statistical analysis demonstrates that contrast-enhanced preprocessing significantly improves defect detection, addressing key limitations in visual inspection. SurgScan provides a scalable, cost-effective AI solution for automated quality control, reducing reliance on manual inspection while ensuring compliance with ISO 13485 and FDA standards, paving the way for enhanced defect detection in medical manufacturing.

Brains learn to represent information from a large set of stimuli, typically by weak supervision. Unsupervised learning is therefore a natural approach for exploring the design of biological neural networks and their computations. Accordingly, redundancy reduction has been suggested as a prominent design principle of neural encoding, but its ``mechanistic'' biological implementation is unclear. Analogously, unsupervised training of artificial neural networks yields internal representations that allow for accurate stimulus classification or decoding, but typically rely on biologically-implausible implementations. We suggest that interactions between parallel subnetworks in the brain may underlie such learning: we present a model of representation learning by ensembles of neural networks, where each network learns to encode stimuli into an abstract representation space by cross-supervising interactions with other networks, for inputs they receive simultaneously or in close temporal proximity. Aiming for biological plausibility, each network has a small ``receptive field'', thus receiving a fixed part of the external input, and the networks do not share weights. We find that for different types of network architectures, and for both visual or neuronal stimuli, these cross-supervising networks learn semantic representations that are easily decodable and that decoding accuracy is comparable to supervised networks -- both at the level of single networks and the ensemble. We further show that performance is optimal for small receptive fields, and that sparse connectivity between networks is nearly as accurate as all-to-all interactions, with far fewer computations. We thus suggest a sparsely interacting collective of cross-supervising networks as an algorithmic framework for representational learning and collective computation in the brain.

Molecular optimization is a central task in drug discovery that requires precise structural reasoning and domain knowledge. While large language models (LLMs) have shown promise in generating high-level editing intentions in natural language, they often struggle to faithfully execute these modifications-particularly when operating on non-intuitive representations like SMILES. We introduce MECo, a framework that bridges reasoning and execution by translating editing actions into executable code. MECo reformulates molecular optimization for LLMs as a cascaded framework: generating human-interpretable editing intentions from a molecule and property goal, followed by translating those intentions into executable structural edits via code generation. Our approach achieves over 98% accuracy in reproducing held-out realistic edits derived from chemical reactions and target-specific compound pairs. On downstream optimization benchmarks spanning physicochemical properties and target activities, MECo substantially improves consistency by 38-86 percentage points to 90%+ and achieves higher success rates over SMILES-based baselines while preserving structural similarity. By aligning intention with execution, MECo enables consistent, controllable and interpretable molecular design, laying the foundation for high-fidelity feedback loops and collaborative human-AI workflows in drug discovery.

We present Natural Language Tools (NLT), a framework that replaces programmatic JSON tool calling in large language models (LLMs) with natural language outputs. By decoupling tool selection from response generation, NLT eliminates task interference and format constraints that degrade tool call performance. When evaluated across 10 models and 6,400 trials spanning customer service and mental health domains, NLT improves tool calling accuracy by 18.4 percentage points while reducing output variance by 70%. Open-weight models see the largest gains, surpassing flagship closed-weight alternatives, with implications for model training in both reinforcement learning and supervised fine-tuning stages. These improvements persist under prompt perturbations and extend tool-calling capabilities to models lacking native support.

While Neural Network pruning typically requires retraining the model to recover pruning-induced performance degradation, state-of-the-art Large Language Models (LLMs) pruning methods instead solve a layer-wise mask selection and reconstruction problem on a small set of calibration data to avoid full retraining, as it is considered computationally infeasible for LLMs. Reconstructing single matrices in isolation has favorable properties, such as convexity of the objective and significantly reduced memory requirements compared to full retraining. In practice, however, reconstruction is often implemented at coarser granularities, e.g., reconstructing a whole transformer block against its dense activations instead of a single matrix. In this work, we study the key design choices when reconstructing or retraining the remaining weights after pruning. We conduct an extensive computational study on state-of-the-art GPT architectures, and report several surprising findings that challenge common intuitions about retraining after pruning. In particular, we observe a free lunch scenario: reconstructing attention and MLP components separately within each transformer block is nearly the most resource-efficient yet achieves the best perplexity. Most importantly, this Pareto-optimal setup achieves better performance than full retraining, despite requiring only a fraction of the memory. Furthermore, we demonstrate that simple and efficient pruning criteria such as Wanda can outperform much more complex approaches when the reconstruction step is properly executed, highlighting its importance. Our findings challenge the narrative that retraining should be avoided at all costs and provide important insights into post-pruning performance recovery for LLMs.

Deep research web agents not only retrieve information from diverse sources such as web environments, files, and multimodal inputs, but more importantly, they need to rigorously analyze and aggregate knowledge for insightful research. However, existing open-source deep research agents predominantly focus on enhancing information-seeking capabilities of web agents to locate specific information, while overlooking the essential need for information aggregation, which would limit their ability to support in-depth research. We propose an Explore to Evolve paradigm to scalably construct verifiable training data for web agents. Begins with proactive online exploration, an agent sources grounded information by exploring the real web. Using the collected evidence, the agent then self-evolves an aggregation program by selecting, composing, and refining operations from 12 high-level logical types to synthesize a verifiable QA pair. This evolution from high-level guidance to concrete operations allowed us to scalably produce WebAggregatorQA, a dataset of 10K samples across 50K websites and 11 domains. Based on an open-source agent framework, SmolAgents, we collect supervised fine-tuning trajectories to develop a series of foundation models, WebAggregator. WebAggregator-8B matches the performance of GPT-4.1, while the 32B variant surpasses GPT-4.1 by more than 10% on GAIA-text and closely approaches Claude-3.7-sonnet. Moreover, given the limited availability of benchmarks that evaluate web agents' information aggregation abilities, we construct a human-annotated evaluation split of WebAggregatorQA as a challenging test set. On this benchmark, Claude-3.7-sonnet only achieves 28%, and GPT-4.1 scores 25.8%. Even when agents manage to retrieve all references, they still struggle on WebAggregatorQA, highlighting the need to strengthen the information aggregation capabilities of web agent foundations.