AI Research news and analysis on Web Pulse: 8281 curated articles tracking the latest AI Research developments, tools, and research, updated continuously from vetted sources.
Researchers introduced SynthAVE, a large-scale benchmark for synthetic labeling of e-commerce attribute extraction, covering 12,726 products across 229 product types, 792 attributes, and four languages. They validated th…
Researchers at an undisclosed institution found that in hierarchical search agents using large language models, scaling the delegation role improves exact match by ~11 points while scaling execution only improves by ~2.6…
Researchers introduced AgentLens, a benchmark for evaluating interactive code agents that assesses entire trajectories—including instruction following, tool use, error recovery, and communication—rather than just binary …
Researchers proposed a ReAct-style agentic setup combining LLM reasoning with SageMath for solving research-level mathematical problems, achieving an average performance gain of 9.7 percentage points across models. GPT-5…
Researchers are developing Agentic Data Environments, an execution substrate that amplifies autonomous agent capabilities while enforcing safety guarantees, reframing data systems from passive stores into active substrat…
Researchers introduced TriRoute, a unified controller that jointly optimizes attention resolution, expert selection, and KV-cache bit-width for language models, achieving Pareto-dominant performance over independent meth…
A new audit of grokking representation metrics reveals that effective rank measurements at the grokking transition overstate converged values by 3-5x on MLPs and 1.3-1.5x on transformers, with compression lagging accurac…
A study on text-to-SQL systems found that verification-based signals, such as LLM judges, outperform black-box signals like self-consistency in predicting query correctness, with a two-provider ensemble achieving 0.82 AU…
Researchers introduced WildCity, a real-world multimodal dataset collected by autonomous fleets traversing complex urban environments, to advance city-scale spatial intelligence for AI. The dataset includes 18 trajectori…
Researchers introduced Gimitest, an open-source framework for testing reinforcement learning policies across diverse environments and algorithms. The tool supports single- and multi-agent RL policies, aiming to improve r…
A new study from arXiv shows that grounding LLM-generated tests in a specification improves code correctness by 38 percentage points over baseline across multiple Claude models, while merely increasing test quantity or c…
Machine learning has fundamentally reoriented scientific practice from inference toward prediction, according to a new analysis of 4.9 million publications and 255 ML techniques by Malena Méndez Isla et al. The study ide…
Researchers at the GRAPES-3 experiment developed a graph neural network (GNN)-based pipeline for cosmic ray energy reconstruction, achieving improved accuracy and automated feature extraction on sparse detector arrays. T…
A new study from arXiv finds that LLM-generated skill files do not improve the performance of AI data scientists across four data-science lifecycle stages. In 7,560 runs covering 56 tasks, nine model configurations, and …
Researchers at arXiv show that biased LLM judges silently disable skill retirement in self-evolving agents, causing a mechanism failure that no amount of data can fix. The study reveals that false-pass bias, where failur…
A comprehensive survey of 1,250 arXiv papers (2024-2026) on AI self-improvement reveals a taxonomy distinguishing bounded self-refinement from open-ended recursive self-improvement (RSI), with the latter constrained by g…
Researchers introduced Pyligent, a training framework that teaches AI models to recover from errors by learning from failed reasoning branches. In tests on hidden graphs, Sudoku, and Blocksworld, Pyligent improved solve …
Researchers have developed BatteryMat, a three-tier machine-learning-to-density-functional-theory (ML-DFT) workflow for screening lithium-ion cathode materials, as detailed in a July 7, 2026 arXiv preprint. The pipeline …
Researchers propose a new evaluation methodology for LLM-powered software engineering agents, focusing on contamination-awareness, in-the-wild behavior assessment, and trajectory-aware benchmarks to better reflect real-w…
A PDF titled 'Practical Lessons from Reinforcement Learning Post Training Experiments' is unavailable due to restricted access from unusual network traffic. The resource's host has blocked the request, citing a security …