Steering Robustness into World Action Models via Mechanistic Interpretability and Optimal Control
This research utilizes mechanistic interpretability to identify and improve the robustness of World Action Models against distribution shifts.
This research utilizes mechanistic interpretability to identify and improve the robustness of World Action Models against distribution shifts.
This paper investigates why large language models struggle with complex context learning and proposes a method using self-discovered specifications to improve performance.
The proposed RAD framework improves offline reinforcement learning by retrieving high-quality demonstrations to help models generalize beyond static training datasets.
Researchers have introduced a new benchmark designed to evaluate how language models handle ambiguous instructions, policy conflicts, and adversarial commands.
The paper presents the NVAITC AI Scientist, a governed end-to-end agentic research system designed to coordinate complex scientific workflows in institutional biomedical environments.
A randomized field experiment evaluated how professional and playful emails rewritten by GPT-5 affected recipient engagement and behavior in workplace communications.
The proposed T2MLR architecture introduces temporal recurrence in middle layers to improve reasoning persistence in transformer models.
Google is integrating its Gemini AI assistant into Waze to enable conversational voice commands for reporting traffic incidents and suggesting map updates.
Researchers introduced Beaver, an agent framework designed to extract and synthesize structured information from complex, multimodal scientific documents.
The GRASP framework improves agentic retrieval-augmented generation by optimizing when to retrieve information and controlling the granularity of the retrieved context.
Chinese entertainment firms are creatively readapting classic cultural intellectual property using modern interactive gaming and live streaming formats.
The UK Ministry of Defence is investing 2 billion pounds to establish an artificial intelligence combat laboratory aimed at upgrading army training and modernizing military exercises.
IBM shares declined significantly following a preliminary report indicating a performance shortfall for the second quarter of 2026.
Apple has increased the retail prices of its iPhone, Apple Watch, and AirPods lineups in Japan, citing factors such as supply chain costs and currency fluctuations.
Sifangda projects a significant increase in net profit for the first half of 2026, driven by growth in its CVD diamond business segment.
Jiaxuan Intelligent, a manufacturer of industrial permanent magnet drive rollers, has filed for an IPO on the Hong Kong Stock Exchange.
China's Ministry of Transport reported a reduction in logistics costs relative to GDP and outlined plans to further optimize transport infrastructure networks during the 15th Five-Year Plan.
RISC-V chip developer Zhichuangxin has completed a Pre-A funding round to advance the integration of open-source hardware and software ecosystems.
Baidu has received board approval to pursue a voluntary conversion to a dual primary listing on the Hong Kong Stock Exchange.
Industry experts are reacting to Sam Altman's skeptical comments regarding the viability of space-based data centers for AI workloads.
A new 1 billion RMB industrial fund focused on advanced materials has been established in Hunan province.
XPeng has launched its MONA L03 smart SUV, which incorporates AI-driven features, for global distribution across 65 markets.
Electric vehicle battery swapping company Aodong New Energy has filed for a Hong Kong IPO amid significant financial losses and declining revenue.
A corporate registration change shows that executive leadership has shifted at a Chinese educational and internet service company.