Real-time AI-curated news from 94,868+ articles across 50+ sources. Sentiment analysis, importance scoring, and key takeaways — updated every 15 minutes.
Researchers propose a learned 'rerooter' approach to improve Levin Tree Search for complex single-agent problems, eliminating the need for explicit subgoal generation. Three rerooter designs exploit state-space structure, learned heuristics, or hybrid signals to achieve scalable search with lower computational overhead and improved online training efficiency.
Researchers introduce GRiD, a novel framework using diffusion models and reinforcement learning to discover complex graph-like rules for knowledge graph reasoning, moving beyond traditional chain-based rule mining. The approach combines supervised pre-training with policy gradient optimization to generate interpretable logical rules while overcoming computational bottlenecks, achieving competitive performance on KG completion benchmarks.
Researchers introduce Adaptive Context Management (AdaCoM), an external LLM-based system that optimizes how AI agents handle long-context tasks by learning agent-specific compression strategies through reinforcement learning. The approach improves performance on web search and research benchmarks while avoiding the need to retrain frozen agents, revealing that high-performing agents benefit from preserving context fidelity while weaker agents need more aggressive compression.
Researchers introduce PReMISE, a framework for auditing and improving rubrics used by LLM judges to evaluate open-ended responses. The work reveals that existing rubrics—whether raw or human-created—fail to simultaneously achieve reliability, preference alignment, and adversarial robustness, with implications for how AI systems measure quality at scale.
Researchers introduce DecomposeR, a framework that trains language models to conduct deep research by explicitly representing plans as directed acyclic graphs rather than flat trajectories. The approach separates planning and execution into two distinct reinforcement learning stages, improving long-form answer generation by 5.1-8.0 points over comparable baselines on benchmark datasets.
Researchers introduce COMPASS, a safety alignment framework for LLM-powered search agents that prevents harmful outcomes from seemingly innocent multi-step queries. The method combines cognitive tree exploration and step-wise alignment to achieve robust safety while maintaining utility, requiring less training data than existing approaches.
UniScale introduces a unified framework that combines model routing and test-time scaling to optimize large language model inference, balancing quality and computational cost. The system uses online learning via contextual multi-armed bandits to adapt inference policies dynamically, achieving fine-grained performance improvements over existing decoupled approaches.
Researchers introduced BilliardPhys-Bench, a benchmark that tests multimodal AI models' ability to predict physical interactions in billiards simulations. The evaluation reveals that leading LLMs from OpenAI, Anthropic, Google, and Alibaba struggle with dynamic physics reasoning, exhibiting systematic failures including a 'stasis bias' where models default to predicting no interaction when physical outcomes become difficult to infer.
Researchers propose a persona-based evaluation framework that replaces traditional monolithic AI benchmarking with diverse synthetic cognitive profiles to better capture cultural and demographic variability in human judgment. While generative models can instantiate these personas consistently, the study reveals systematic degradation in persona coherence over time, suggesting static alignment approaches are insufficient and dynamic regulatory mechanisms are needed.
Researchers propose HADT, a transformer-based AI architecture designed to optimize autonomous resource management in heterogeneous satellite clusters conducting Earth Observation missions. The model-free reinforcement learning approach replaces traditional mathematical optimization methods, demonstrating improved performance and adaptability across varying satellite configurations.
Researchers introduce GraphARC, a new benchmark for evaluating artificial intelligence systems on abstract reasoning tasks using graph-structured data. The framework extends the popular ARC benchmark to graph domains, revealing significant limitations in current language models—particularly a gap between understanding graph properties and executing complex transformations, with performance degrading substantially on larger instances.
Researchers present a novel technique for matching vectors across different AI embedding models trained independently on overlapping datasets. The method leverages local geometric consistency in contrastive encoders to establish cross-model correspondences using only a small seed set of paired anchors, with applications to vector database integration.
Researchers formalize causal pathway analysis for rare events in structural equation models, proposing testable implications that depend on causal abstractions rather than complete system graphs. This work bridges verbal explanations and rigorous causal modeling, enabling root cause analysis of outliers with reduced computational complexity.
Researchers introduce GLIDE, an open-source Python library that standardizes prediction-powered inference (PPI) methods for evaluating AI systems and language models. The library combines human annotation with LLM evaluations to produce unbiased estimates with valid confidence intervals, potentially reducing annotation costs while maintaining accuracy.
TraceGraph is a new graph-based framework that analyzes multi-model agent trajectories to create shared decision landscapes, revealing how different AI models navigate tasks differently. The tool identifies failure regions and trap states, enabling targeted improvements that increased resolved rates on SWE-bench by 3-4.8%, demonstrating that aggregate benchmark scores mask critical performance divergences.
Researchers introduce CoSee, an auditing framework for analyzing failure modes in collaborative visual reasoning systems using resource-constrained language models (4B-8B parameters). The study reveals that shared working memory architectures paradoxically amplify hallucinations rather than improve performance, identifying two critical failure modes: noise reinforcement and policy collapse.
Researchers introduce SCALE, a self-improving web agent framework that uses adversarial roles and cognitive-aware exploration to autonomously adapt to complex web environments without relying on handcrafted pipelines or expensive expert data. The framework includes SCALE-Hop, a graph exploration strategy, and SCALE-20k, a 20,000-sample dataset from 19 real-world websites that demonstrates improved performance across multiple multimodal large language models.
HypoAgent is a new AI framework that uses multiple specialized agents to generate logical hypotheses from knowledge graphs through interactive dialogue. The system excels at understanding evolving user intent across multi-turn conversations and diagnosing why generated hypotheses fail, achieving state-of-the-art performance on both commonsense and biomedical knowledge graphs.
Researchers introduce FAM-Bench, a multimodal benchmark dataset containing 2,500 expert-verified instances designed to evaluate AI models' ability to assess food suitability for specific health conditions. The benchmark addresses a gap in existing food AI systems by testing health-aware reasoning through dish suitability assessment and comparative analysis tasks across 13 diet-related conditions.
Researchers have developed an Answer-Set Programming (ASP) based implementation of the CARCASS framework to improve Reinforcement Learning abstractions for complex state spaces. The approach leverages ASP's declarative modeling capabilities as an alternative to Prolog, demonstrating promising results in Blocks World and Minigrid domains when domain knowledge is available.
Researchers introduce AutoSci, an AI-driven system designed to automate the full scientific research lifecycle by managing literature review, experiments, manuscript writing, and peer review responses. The system uses a memory-centric architecture with four specialized modules to maintain structured knowledge, execute research workflows, and continuously improve its procedures through feedback.
Researchers demonstrate that Large Language Models improve their reasoning performance when search histories are explicitly structured with parent pointers (LinTree), rather than implicitly represented. The finding suggests that LLMs benefit from tree-aware representations during problem-solving, outperforming both implicit trace-based reasoning and traditional heuristic-guided search across multiple domains.
Researchers introduce context-dependent argumentation frameworks (CDAFs) extending Dung's argumentation theory to capture strategic manipulation of argument validity across different contexts. The framework models how an agent can selectively activate relevant criteria to influence which arguments succeed, introducing a new decision problem called ACTIVATION-MANIPULATION with unexplored complexity bounds.
Researchers demonstrate that LLM-generated reward functions for reinforcement learning tasks fail in predictable ways and are better treated as an iterative debugging process rather than one-shot generation. Using diagnostic-driven refinement guided by failure-mode taxonomy, they improve task success rates significantly (DoorKey-8x8: 2.3% to 97.6%), though the method shows limitations in dense-reward continuous control and requires reliable semantic interfaces.
Researchers introduce EGGROLL, a low-rank factorization technique that enables gradient-free training of Spiking Neural Networks (SNNs) using Evolution Strategies, reducing computational overhead by 2.23x while maintaining 79.21% accuracy on N-MNIST. This breakthrough addresses the long-standing challenge of training SNNs on neuromorphic hardware without requiring backpropagation infrastructure.
