Brief

New research platforms like OpenG2G are being developed to simulate and coordinate AI datacenters with the electricity grid, addressing challenges like interconnection delays and power flexibility. Simultaneously, scalable digital twin frameworks are emerging to optimize energy consumption within datacenters using predictive models. These advancements come as AI's immense power demands strain existing infrastructure, prompting discussions on co-design principles and innovative power architectures to meet future needs. AI

Chinese President Xi Jinping met with US President Donald Trump in Beijing for a state visit, aiming to stabilize bilateral relations and foster cooperation. The leaders discussed overcoming the "Thucydides Trap" and finding a path for major powers to coexist peacefully. Preceding the summit, high-level trade talks were held in Seoul, with both sides seeking to resolve economic issues and potentially reach quick agreements. The visit also included opportunities for networking between US and Chinese business leaders. AI

A new paper introduces a stateful transformer inference engine that significantly speeds up processing for streaming data by maintaining a persistent KV cache. This approach allows for query latency that is independent of accumulated context size, achieving up to a 5.9x speedup on market-data benchmarks compared to existing engines. Separately, Intel has released AutoRound, an advanced quantization toolkit for LLMs and VLMs that enables high accuracy at ultra-low bit widths (2-4 bits) with broad hardware compatibility, integrating with popular frameworks like vLLM and Transformers. AI

Researchers are exploring new methods to enhance AI safety and efficiency. One paper proposes a language-agnostic approach to detect malicious prompts by comparing query embeddings against a fixed English codebook of jailbreak prompts, showing promise but also limitations under distribution shifts. Another study investigates how the wording of schema keys in structured generation tasks can implicitly guide large language models, revealing that different models like Qwen and Llama respond differently to prompt-level versus schema-level instructions. Separately, a discussion highlights the increasing importance and evolving landscape of open-weights models, noting that while they offer cost and privacy advantages, their availability and licensing are becoming more restrictive. AI

A new benchmark, AmBench, reveals that large language models struggle to reliably recognize human names, a critical component for privacy protection tools. Researchers found that LLMs mishandle ambiguous names, leading to a 20-40% drop in recall compared to more recognizable names. This uneven privacy protection raises fairness concerns, particularly when prompt injections cause LLMs to ignore names, as seen in Anthropic's Clio tool. AI

Hugging Face has announced several updates and collaborations across its platform. These include enhancements to OCR pipelines with open models, the integration of Sentence Transformers, and the release of Transformers.js v4. Additionally, Hugging Face is strengthening AI security through a partnership with VirusTotal and introducing new models like Granite 4.0 Nano and AnyLanguageModel for efficient LLM operations. AI

ChatGPT has reportedly outperformed human applicants on the 2026 entrance exams for the University of Tokyo and Kyoto University, a significant leap from GPT-4's performance two years prior. Meanwhile, OpenAI is testing a self-service ad manager for ChatGPT, with plans to roll it out in Japan. Anthropic has introduced a "Dreaming" feature for its Claude Managed Agents, enabling them to learn from past failures and potentially develop more sophisticated autonomous behaviors. AI

Apple has released iOS 26.5, which includes end-to-end encrypted RCS messaging in beta, a feature developed in collaboration with Google. This update aims to enhance the privacy and security of cross-platform text messages between iPhones and Android devices, a significant change for Google Messages. While Apple emphasizes iMessage's continued superiority for inter-device communication, the RCS encryption is a major step for Android users seeking secure messaging options beyond WhatsApp. The rollout is dependent on carrier support and may not be immediately available to all users. AI

Several trade shows are scheduled in China for 2026, focusing on artificial intelligence and related technologies. The Guangzhou International Smart Equipment and Artificial Intelligence Exhibition will take place from June 3-5, 2026, highlighting smart equipment, AI, and robotics. In Shanghai, the AI-Driven Industry Conference & Expo is set for May 28-29, 2026, exploring the intersection of automotive, data centers, and intelligent robotics. Additionally, Tech Week Shanghai will occur on May 6-7, 2026, emphasizing data industrialization and AI infrastructure. AI

Security researchers have identified significant vulnerabilities in several Model Context Protocol (MCP) servers, including those from Atlassian, GitHub, Cloudflare, and Microsoft. The most common critical flaw is indirect prompt injection, where attackers can manipulate data fetched by MCP servers to trick AI agents into executing malicious instructions. Other issues include privilege escalation through mislabeled tool permissions and Server-Side Request Forgery (SSRF) vulnerabilities in HTTP-calling tools. These findings highlight a substantial security risk in the MCP ecosystem, with nearly 30% of scanned packages exhibiting high or critical severity vulnerabilities. AI

A Chinese-American man, Lu Jianwang, has been found guilty in New York of acting as an unauthorized agent of China and obstruction of justice. Prosecutors alleged that Lu operated a secret Chinese police station in Manhattan's Chinatown, which was used to monitor and harass dissidents. His defense team argued that the location was merely a community center for overseas Chinese to renew driver's licenses and engage in social activities. AI

Owners of Hong Kong's fire-ravaged Wang Fuk Court are pushing back against the government-appointed administrator's attempt to delay a crucial meeting. Over 240 residents, representing more than 12% of households, petitioned for an extraordinary general meeting to discuss the future of their flats, including a potential sale. The administrator, Hop On Management Company, has requested a delay, citing insufficient time for preparation and the need to verify signatures, while residents threaten legal action if the meeting is not held within the mandated 45-day period. AI

Researchers have developed a new open-source machine learning compiler stack written in just 5,000 lines of Python. This stack offers unprecedented transparency by lowering large language models to CUDA with six intermediate representations. It aims to be hackable and CUDA-optimized, contrasting with more complex systems like PyTorch or TVM. Additionally, AI agents are being highlighted for their potential to automate exploratory data analysis and data preparation tasks, promising significant time savings for data scientists. AI

Meta CEO Mark Zuckerberg announced a new AI

Republicans have temporarily halted aggressive redistricting plans across several Southern states, offering a reprieve to Black lawmakers who faced potential displacement. While Mississippi and Georgia have postponed map redraws, Louisiana will eliminate only one Black-majority district instead of two. Alabama and South Carolina, however, are proceeding with plans that could affect the districts of prominent Black representatives. AI

Gan & Lee Pharmaceuticals has received marketing authorization from the European Commission for its insulin lispro injection, Bysumlog®. This rapid-acting insulin analog is approved for treating diabetes in adults and children across the EU, Iceland, Liechtenstein, and Norway. The company has invested approximately 414 million yuan in its research and development. AI

Hong Kong authorities have formally issued buy-back offer letters to homeowners of the fire-damaged Wang Fuk Court. The government has allocated HK$6.8 billion (US$870 million) to acquire the estate, with offers set at HK$8,000 or HK$10,500 per square foot. Homeowners have until August 31 to accept the offer, with expedited payouts promised for those choosing cash compensation. AI

Hong Kong's Transport Department has postponed plans to lower the mandatory medical certification age for taxi and commercial vehicle drivers from 70 to 65. The proposed changes, which would also require annual check-ups for license renewals, are now anticipated in the latter half of the year. This delay follows recent accidents involving elderly drivers, including one that resulted in a fatality and multiple injuries, prompting concerns about driver health and road safety. AI

Senator Peter Welch has expressed his support for former President Trump's "most-favored nation" drug pricing policy, indicating he would actively work to pass such legislation. This bipartisan backing is seen as crucial for codifying Trump's plan into law, especially as many Americans worry about affording prescription drugs. Welch is already collaborating with Senator Josh Hawley on a similar bill that aims to prevent drug manufacturers from charging higher prices in the U.S. compared to international averages. AI

The US has reiterated an offer of US$100 million in aid to Cuba, to be distributed through independent humanitarian organizations like the Catholic Church. This offer comes as former President Trump reportedly discussed Cuba with Chinese President Xi. The US State Department stated that Cuba's government must decide whether to accept the assistance, accusing it of previously blocking similar aid, including satellite internet support. AI

Israeli Prime Minister Benjamin Netanyahu reportedly made a secret visit to the United Arab Emirates on March 26th. During this meeting with Emirati President Sheikh Mohammed bin Zayed, they discussed a "historic breakthrough" in bilateral relations. This visit occurred during the recent conflict with Iran and signifies a strengthening of ties established by the 2020 Abraham Accords, with the UAE viewing its relationship with Israel as a strategic asset. AI

Senator Ronald dela Rosa, a former Philippine police chief, is facing arrest on charges of crimes against humanity related to former President Rodrigo Duterte's drug war. Gunshots were heard at the Philippine Senate as security forces reportedly entered the building in anticipation of his arrest. Dela Rosa has denied involvement in illegal killings and had urged supporters to mobilize to prevent his handover to the International Criminal Court (ICC). The ICC unsealed an arrest warrant for him, mirroring charges against Duterte. AI

UK Prime Minister Rishi Sunak is facing a significant political challenge as reports suggest Health Minister Wes Streeting is preparing to resign and launch a leadership bid. This potential resignation threat looms over the government's agenda, which was outlined in the King's Speech. Financial markets reacted negatively to the news, with British government bond futures falling and the pound weakening against the euro. AI

Louisiana senators have advanced a new congressional map that reduces the number of majority-Black districts from two to one. This map, proposed by Sen. Jay Morris, pits two incumbent Black representatives against each other for a single seat. An alternative map that would have preserved two "opportunity" districts for Black voters, supported by residents, failed to advance. AI

Malaysia is considering legal action against Norway after Oslo blocked the export of Naval Strike Missiles for Malaysia's troubled littoral combat ship program. Prime Minister Anwar Ibrahim has discussed the issue with his Norwegian counterpart, expressing deep disappointment and rejecting the reasons provided for the export denial. The contract, signed in 2018, had proceeded without issue until this last-minute decision, which Malaysia views as a compromise to its national defense strategy. AI

British Prime Minister Keir Starmer is facing significant pressure to resign following a wave of discontent within his Labour Party, with over a fifth of lawmakers reportedly urging him to step down. Health Secretary Wes Streeting is rumored to be preparing a leadership bid, potentially as early as Thursday, amidst the political turmoil. The crisis comes as King Charles III outlined the government's legislative agenda, a traditional event now overshadowed by internal party struggles and recent electoral losses. AI

South Korea is updating its national weather warning system for the first time in 18 years, introducing a new top-tier "extreme heat emergency" alert. This change reflects a significant increase in heatwave days and tropical nights over the past five years. The new alert will be activated when perceived temperatures reach 38°C or actual temperatures hit 39°C, signaling a serious threat to public health and infrastructure. AI

Senate Republicans are expressing concern over a $220 million request to enhance security at the White House complex, which includes funding for President Trump's new East Wing ballroom. This funding is part of a larger $72 billion package for ICE and Border Patrol, and GOP senators are hesitant due to deficit concerns. Democrats are also opposing the funding, urging it be redirected to public safety initiatives instead of the ballroom project. AI

Researchers are developing a science of scaling AI agent systems, moving beyond the heuristic that more agents are always better. New studies reveal that multi-agent coordination significantly improves performance on parallelizable tasks but can degrade it on sequential ones. Efforts are underway to create predictive models for optimal agent architecture and to develop methods for real-time evaluation and error mitigation in agent interactions. AI

Hugging Face has released a suite of resources and models focused on advancing vision-language models (VLMs). These include new open-source models like Google's PaliGemma and PaliGemma 2, Microsoft's Florence-2, and Hugging Face's own Idefics2 and SmolVLM. The platform also offers guides and tools for aligning VLMs, such as TRL and preference optimization techniques, aiming to improve their capabilities and accessibility for the community. AI

Anthropic has introduced Natural Language Autoencoders (NLAs), a new method that translates the internal numerical 'thoughts' (activations) of large language models into human-readable text. This technique allows researchers to better understand model behavior, including identifying instances where models might be aware of being tested but do not verbalize it, or uncovering hidden motivations. While NLAs offer a significant advancement in AI interpretability and debugging, Anthropic notes limitations such as potential 'hallucinations' in the explanations and high computational costs, though they are releasing the code and an interactive frontend to encourage further research. AI

Google Research has developed a framework to evaluate the alignment of Large Language Models (LLMs) with human behavioral dispositions, using established psychological assessments adapted into situational judgment tests. This approach quantizes model tendencies against human social inclinations, identifying deviations and areas for improvement in realistic scenarios. Separately, Google Research also introduced SLED (Self Logits Evolution Decoding), a novel method that enhances LLM factuality by utilizing all model layers during the decoding process, thereby reducing hallucinations without external data or fine-tuning. AI

Microsoft has developed its own custom AI chips, the Azure Maia 100 AI accelerator and the Azure Cobalt 100 CPU, to power its Azure cloud infrastructure. These in-house designed chips aim to reduce reliance on third-party providers like Nvidia and optimize performance and cost for AI workloads, including training and inference for large language models. The Maia chip is being developed in collaboration with OpenAI, with CEO Sam Altman highlighting its potential to make model training more capable and affordable. AI

Researchers are developing novel methods to combat hallucinations in Large Language Models (LLMs). Several papers propose new frameworks and techniques, including LaaB, which bridges neural features and symbolic judgments, and CuraView, a multi-agent system for medical hallucination detection using GraphRAG. Other approaches focus on neuro-symbolic agents for hallucination-free requirements reuse, adaptive unlearning for surgical hallucination suppression in code generation, and harnessing reasoning trajectories via answer-agreement representation shaping. Additionally, new benchmarks like HalluScan are being created to systematically evaluate detection and mitigation strategies. AI

Recent research explores novel methods to enhance the reasoning capabilities and efficiency of large language models (LLMs). Papers introduce techniques like speculative exploration for Tree-of-Thought reasoning to break synchronization bottlenecks and achieve significant speedups. Other work focuses on improving tool-integrated reasoning by pruning erroneous tool calls at inference time and developing frameworks for robots to perform physical reasoning in latent spaces before acting. Additionally, research investigates the effectiveness of different reasoning protocols, such as debate and voting, for LLMs, finding that while some methods improve safety, they don't always enhance usefulness. AI

Recent research explores advancements in 3D Gaussian Splatting (3DGS), a technique for real-time photorealistic novel-view synthesis. New methods like GETA-3DGS focus on efficient compression through joint pruning and quantization, achieving significant storage reduction. Other work, such as EnerGS, introduces soft geometric guidance to improve reconstruction quality in challenging outdoor scenarios with incomplete data. Additionally, FreeTimeGS++ enhances dynamic scene reconstruction by analyzing and optimizing temporal partitioning and spatiotemporal consistency, while WildSplatter enables feed-forward 3DGS from unconstrained images with appearance control. AI

Apple's research paper explores the mechanisms behind compositional generalization in conditional diffusion models, specifically focusing on how they handle combinations of conditions not seen during training. The study validates that models exhibiting local conditional scores are better at generalizing, and that enforcing this locality can improve performance. Separately, Hugging Face has released several blog posts detailing various methods for fine-tuning and optimizing Stable Diffusion models, including techniques like DDPO, LoRA, and optimizations for Intel CPUs, as well as instruction-tuning and Japanese language support. AI

OpenAI has published a series of research papers detailing advancements in reinforcement learning (RL). These include achieving superhuman performance in the game Dota 2 using large-scale deep RL, developing benchmarks for safe exploration in RL environments, and quantifying generalization capabilities with a new environment called CoinRun. The research also explores novel methods like Random Network Distillation for curiosity-driven exploration, Evolved Policy Gradients for faster learning on new tasks, and variance reduction techniques for policy gradients. Additionally, OpenAI is investigating policy representations in multiagent systems and the theoretical equivalence between policy gradients and soft Q-learning. AI

Google DeepMind has introduced the FACTS Benchmark Suite, a new set of evaluations designed to systematically assess the factuality of large language models across various use cases. This suite includes benchmarks for parametric knowledge, search-based information retrieval, and multimodal understanding, alongside an updated grounding benchmark. The initiative aims to provide a more comprehensive measure of LLM accuracy and is being launched with a public leaderboard on Kaggle to track progress across leading models. AI

Anthropic conducted an experiment where Claude agents acted as digital barterers, successfully negotiating 186 deals totaling over $4,000. Participants found the deals fair, with nearly half expressing willingness to pay for such a service. The experiment highlighted that while model quality, such as Opus versus Haiku, significantly impacted deal outcomes, human participants did not perceive this difference. AI