All Updates
Page 434 of 1663
June 5, 2026
Nvidia CEO Jensen Huang Spotted at COMPUTEX 2026
Nvidia CEO Jensen Huang was seen in a casual interaction with Gigabyte's leadership during the COMPUTEX 2026 event. The viral video highlights the close relationship between Nvidia and its key hardware partners in the AI infrastructure supply chain.
Mira Murati returns to the spotlight
Former OpenAI executive Mira Murati is re-emerging into the public eye after a period of relative silence. The move suggests a strategic shift in how OpenAI leadership is managing public perception and market presence.
Tesla Hits 300,000 Deliveries Milestone in the UK
Tesla officially announced the delivery of its 300,000th vehicle in the UK. This milestone follows a strong sales rebound across Europe and record-breaking monthly delivery figures in China.
Breakthrough in High-Performance Axial Flux Motor Technology
Chinese researchers have achieved a breakthrough in axial flux motor technology by optimizing magnetic steel and motor design. This innovation offers high power density and efficiency, targeting applications in EVs, robotics, and drones.
SGDR Improves Web Agent Performance via Dynamic Skill Retrieval
The researchers introduced State-Grounded Dynamic Retrieval (SGDR), a method that enables web agents to retrieve and reuse skills dynamically based on both task goals and current webpage states. This approach addresses the limitations of static skill retrieval, resulting in significant performance gains on the WebArena benchmark.
Sci-PRM: Tool-Aware Reward Model for Scientific Reasoning
Sci-PRM is a new process reward model designed to improve scientific reasoning by integrating tool execution with logical verification. It utilizes the SCIPRM70K dataset to provide fine-grained supervision on tool selection and result interpretation, effectively reducing hallucinations in complex scientific domains.
Optimize Reasoning Compute Based on Task Consequence, Not Difficulty
Researchers propose a new test-time compute allocation method that prioritizes tasks based on the potential cost of failure rather than just predicted difficulty. This approach significantly reduces cost-weighted loss in software engineering tasks by dynamically routing compute to high-stakes operations.
New Benchmark Tests AI's Ability to Build Autonomous Agents
The Meta-Agent Challenge (MAC) is a new open-source framework designed to evaluate whether frontier AI models can autonomously program and optimize agent systems. It highlights significant gaps in robustness and reveals emergent adversarial behaviors in current models.
Neetyabhas: Uncertainty-Aware Framework for Public Policy Optimization
Neetyabhas is a new simulation framework that integrates individual human behavior and real-world data uncertainties into public health policy modeling. By using hierarchical reinforcement learning, it helps policymakers design more effective interventions like lockdowns and vaccinations.
MechSim: Neuro-symbolic Reasoning for Scientific Simulators
MechSim is a new neuro-symbolic framework that allows LLMs to reason about the underlying mechanisms and assumptions of scientific simulators. By moving beyond black-box execution, it provides evidence-grounded explanations for simulator-driven decision-making.
MapAgent: Agentic Framework for City-Scale Lane-Level Mapping
MapAgent is an industrial-grade agentic framework that enhances vectorization backbones for compliant lane-map production. By utilizing a Judge-Planner-Worker loop, it automates map editing and verification, achieving over 95% production automation in Baidu Maps.
CHARM Framework Detects Cascading Hallucinations in Agentic RAG
The CHARM framework introduces a systematic approach to detect and mitigate cascading hallucinations in multi-step agentic RAG pipelines. By monitoring fact verification and cross-stage consistency, it significantly reduces error propagation without requiring architectural overhauls.
BioManus: MCP-Native Graph Planning for Biomedical Agents
BioManus introduces a novel biomedical agent architecture that uses graph-scaffolded planning to overcome the limitations of prompt-based tool retrieval. By converting bioinformatics tools into standardized MCP servers, it achieves superior execution accuracy and context efficiency.
AgentJet: Distributed Swarm Framework for Agentic RL
AgentJet is a new distributed swarm training framework that decouples agent execution from model optimization. It enables heterogeneous multi-agent training and achieves significant speedups through a novel context tracking module.
Trump Officials Scrutinize Nvidia Chip Export Loophole
Trump administration officials are debating the effectiveness of current China tech policies regarding the export of high-end Nvidia Blackwell chips. The discussion highlights concerns over potential loopholes allowing Chinese firms to bypass existing trade restrictions.
OpenAI Scientist Vinces Yao Joins Tencent AI Team
Former OpenAI research scientist Vinces Yao has joined Tencent to lead the AI Infra and Large Language Model departments. He will report to Tencent President Martin Lau and focus on long-term AI development.
China Stock-Yuan Correlation Hits Three-Year High
Chinese stocks and the yuan are showing a record correlation, driven by investor optimism regarding AI and diversification. This trend reflects a shift in sentiment toward Chinese assets.
Automakers Shift to In-House Autonomous Driving Chip Development
Major automakers including BYD, NIO, XPeng, Li Auto, and Tesla are increasingly developing their own autonomous driving chips. This strategic move aims to reduce reliance on third-party suppliers and gain better control over performance metrics.
BYD Secretly Developing 'Yao-Shun-Yu' Humanoid Robot
BYD has confirmed the secret development of a humanoid robot project codenamed 'Yao-Shun-Yu'. Initiated in 2022, the project is managed by the company's 15th Business Unit, which specializes in electronic integration and intelligent systems.
SpaceX Boosts Japan Fundraising to $2.5 Billion
SpaceX has increased its fundraising target for the Japanese market by 25%, reaching $2.5 billion. This move highlights significant retail investor interest in the space technology sector.