Pulse

Chinese President Xi Jinping met with former U.S. President Donald Trump in Beijing, aiming to stabilize bilateral relations and find a path for cooperation despite ongoing strategic rivalry. The leaders discussed critical issues including Taiwan, trade, and the global economy, with both sides expressing a desire for partnership over confrontation. While the summit was described as historic and productive, underlying tensions and competition, particularly concerning technology and geopolitical influence, remain significant challenges for future U.S.-China interactions. AI

Immigration and Customs Enforcement (ICE) is leveraging Palantir's technology to enhance its ability to locate individuals for arrest and raid locations. A senior ICE official revealed that agents now have access to a list of 20 million potential targets on their iPhones, significantly increasing the speed and success rate of operations. This system, which integrates data from numerous sources, has reportedly reduced the time for target identification from hours to minutes. 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

A recent study revealed significant inconsistencies in AI models' ability to accurately estimate carbohydrate counts from food images, posing potential health risks for diabetes management. Across over 26,000 queries, models like OpenAI's GPT-5.4 and Google's Gemini 2.5 Pro exhibited wide variations in carb estimates for the same meal, with some discrepancies large enough to necessitate dangerous insulin dosing errors. The research also highlighted instances where models misidentified foods or hallucinated ingredients, further compromising their reliability for critical health applications. AI

The Authors Alliance has released a final report and outlined a path forward for the Public Interest Corpus. This initiative focuses on the ethical considerations and practicalities of using digital books for AI training. The report emphasizes the need for responsible data practices within the AI development community, particularly concerning copyrighted materials. AI

AI models, particularly GPT-5.5, have exhibited a peculiar behavior dubbed "goblin mode," characterized by an unusual fixation on goblin-related imagery and language. This phenomenon gained traction on AI Twitter, with users experimenting and sharing observations. While some speculate it's an artifact of RLHF training or a quirky response to coding prompts, direct attempts to replicate the behavior under controlled conditions have yielded mixed results, suggesting it may not be as easily elicited as initially believed. AI

A critical security vulnerability (CVE-2026-5760) with a severity score of 9.8 has been identified in SGLang, an AI inference server. The issue arises from a poisoned GGUF model file containing a chat-template that SGLang processes via an unsandboxed Jinja2, allowing arbitrary Python code execution on the host system. This vulnerability is similar to past issues found in llama-cpp-python and vLLM, highlighting a persistent oversight in handling model file templates across multiple AI frameworks. AI

An effort by manufacturers to repeal Colorado's right-to-repair law has been unsuccessful. The law, which allows consumers and independent repair shops to access parts and manuals for electronic devices, faced a repeal attempt that ultimately failed. This outcome preserves the existing consumer protection measures in the state. AI

A new arXiv preprint titled "Borges and AI" explores the intersection of artificial intelligence and the literary works of Jorge Luis Borges. The paper, authored by Léon Bottou and Bernhard Schölkopf, delves into philosophical and conceptual connections between Borges's writings and AI. It examines how themes present in Borges's fiction, such as labyrinths, infinite libraries, and the nature of reality, can be interpreted through the lens of modern AI. AI

A recent survey by BookNet Canada and BISG reveals that nearly half of the publishing industry is currently utilizing AI technologies. Despite this widespread adoption, the qualitative feedback indicates a predominantly negative sentiment towards AI within the sector. The report suggests that the publishing industry's embrace of AI is outpacing its comfort level with the technology. AI

Researchers at Georgia Tech have developed a free online tool called the Transformer Explainer to help non-experts understand how large language models like ChatGPT and Claude process language. Separately, new research from the University of British Columbia, presented at the 2026 CHI Conference, suggests that the design of AI chatbots may contribute to addictive usage patterns among some users. AI

The META/e Conference on Transparency, Technology and AI in Peer Review is set to explore the integration of AI and technology within academic processes. Discussions will center on how these tools can be applied to peer review, the evaluation of research grants, and broader research assessment methodologies. The event aims to foster dialogue on the evolving landscape of academic evaluation. AI

ThinkPub has released a new report examining the European translation market, prompted by the cessation of UNESCO's Index Translationum. The report, titled "The Importance of Translation for the Public Good," delves into market dynamics and industry statistics within Europe. This initiative aims to fill the void left by the discontinuation of the Index Translationum, a key resource for tracking translated works. AI

New York City Schools Chancellor David C. Banks has withdrawn controversial plans that included opening the city's first high school dedicated to artificial intelligence. The proposals also involved closing two middle schools and relocating another. This decision follows public outcry and concerns regarding the focus on AI in education. 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

A recent study, inspired by the Dead Internet Theory, has found that approximately one-third of all newly created websites are generated using AI. This research examines the significant impact that large language models like ChatGPT and their competitors have had on the internet's landscape since 2022. The findings suggest a substantial shift in online content creation, with AI playing an increasingly dominant role. AI

Global tropical rainforest loss decreased in the past year, offering a glimmer of hope in conservation efforts. However, scientists caution that the rate of disappearance remains alarmingly high. The potential impact of El Niño-driven fires poses a significant threat to this progress, highlighting the ongoing challenges in protecting these vital ecosystems. AI

The Center for Internet Security (CIS) has released three new guides focused on AI security. These companion documents address the specific security considerations for Large Language Models (LLMs), autonomous agents, and Multi-Cloud Platform (MCP) environments. The guides are designed to offer practical control mapping for enterprise teams, moving beyond abstract policy discussions. AI

UNESCO and the University of Oxford have launched a new global course titled "AI, Justice and the Rule of Law." This program is designed to train judges and judicial professionals on the responsible use of artificial intelligence. The curriculum aims to ensure AI applications align with human rights and legal principles. AI

College professors are increasingly adopting oral examinations as a strategy to combat the challenges posed by artificial intelligence in higher education. This resurgence of an ancient testing method, combined with other traditional and modern approaches, aims to address a crisis within the academic system. The shift reflects a growing concern among educators about maintaining academic integrity and assessing genuine student understanding in the era of AI. AI

A study involving 26,904 queries across four leading AI models—OpenAI's GPT-5.4, Anthropic's Claude Sonnet 4.6, and two Google Gemini versions—revealed significant inconsistencies in carbohydrate estimations from food images. Despite using the same prompts and images repeatedly, the models produced widely varying results, with some estimations differing by hundreds of grams, potentially leading to dangerous insulin dosing errors for diabetes patients. The study also highlighted instances where models misidentified foods or hallucinated ingredients, further compromising their reliability for health applications. AI

Microsoft has released the source code for 86-DOS 1.00 and early versions of PC-DOS 1.00, making historical operating system code available to the public. This release includes the 86-DOS kernel, development snapshots of PC-DOS, and utilities like CHKDSK, along with the assembler itself. The code was transcribed from printouts and notes found in Tim Paterson's garage, offering a unique glimpse into software development practices from the late 1970s and early 1980s. AI

Anthropic's Claude AI is expanding its capabilities into creative and scientific domains. New connectors allow Claude to interact with 3D modeling in Blender, automate Adobe Creative Cloud tasks, and assist with music production in Ableton. Additionally, Claude has achieved expert-level performance in bioinformatics, matching human specialists on the BioMysteryBench benchmark for complex genomic analysis. AI

OpenAI has released an open-source tool called Privacy Filter 2026, a 1.5 billion parameter model designed to detect and remove personally identifiable information (PII) directly within a user's browser. This approach allows organizations to anonymize text without transmitting sensitive data to external servers, enhancing data privacy. Separately, Meta FAIR has introduced NeuralSet, a Python package that integrates various neuroscience data modalities with AI models, facilitating cross-domain research. AI

Alibaba's Qwen team has released FlashQLA, a new set of high-performance linear attention kernels developed using TileLang. These kernels are designed to improve the efficiency of attention mechanisms in large language models. The team also shared benchmark results for their Qwen models, showcasing performance across various configurations. AI

A perspective was shared suggesting that in overparameterized models, increasing the number of parameters allows for more diverse fitting, enabling the learning of latent structures not found during training. This concept was illustrated using examples of logit models and kernel high-dimensional projections of Support Vector Machines (SVMs). The discussion aimed to provide intuition for the generalization capabilities of larger models. AI

Meta's Fundamental AI Research (FAIR) team has introduced NeuralSet, a new Python package designed to integrate neuroscience data with artificial intelligence models. This tool is capable of processing various neuroimaging and electrophysiology data types, including fMRI, M/EEG, and spike recordings. NeuralSet also supports embeddings from HuggingFace, facilitating a more comprehensive approach to neuro-AI research. AI

A developer shared their experience building and releasing a sales forecasting web application using FastAPI, Prophet, and React. Separately, another article discussed a six-day period where GitHub faced challenges related to governance, uptime, security, and costs. AI

Researchers are developing methods to detect AI-generated text by analyzing network traffic patterns. One approach involves examining the unique digital fingerprints left by AI models during their operation. This could help identify the origin of AI-generated content and potentially mitigate misuse. AI

Researchers at MIT have developed a novel approach to handling prompts in AI models. Instead of placing the prompt within the standard context window, this new method treats the prompt as a separate environment. This innovation aims to potentially improve how AI models process and respond to user inputs. AI

OpenAI has proposed a five-point strategy to enhance cybersecurity during the current Intelligence Age. The plan emphasizes democratizing AI tools for cyber defense and fortifying critical infrastructure. This initiative aims to leverage AI's capabilities to better protect against evolving cyber threats. AI

The National Privacy Commission (NPC) in the Philippines has issued a warning about a new AI

The Data Science Learning Community (DSLC) recently hosted two video sessions focused on machine learning concepts. One session, titled "Deep Learning with Python (3e): Interpreting what ConvNets learn," explored how convolutional neural networks process information. Another session, "An Introduction to Statistical Learning with Python: Unsupervised Learning," provided an overview of unsupervised learning techniques using Python. AI

An AI model named Claude Mythos has identified an unprecedented 271 zero-day vulnerabilities within the Firefox browser. The Firefox security team has been actively addressing these issues since February. This discovery highlights the potential of AI in uncovering complex security flaws. AI

OpenClaw, an open-source autonomous AI agent, has rapidly gained popularity since its release in late 2025. Developed by Peter Steinberger, who later joined OpenAI, the agent utilizes large language models and messaging platforms for task execution. Its architecture and capabilities are being explored in a series of articles, with a focus on its potential to integrate AI into everyday routines and workflows. AI

The Unitree Robotics H1 humanoid robot has demonstrated a sprint speed of 10 meters per second in recent tests conducted in China. The company's CEO anticipates that these robots will surpass the 10-second mark for a 100-meter dash by mid-2026. This advancement highlights significant progress in the field of humanoid robotics and its potential for future applications. AI

MITRE has identified increasing cybersecurity threats associated with the integration of AI, cloud computing, and post-quantum technologies into medical devices. These advancements, while offering potential benefits, introduce new vulnerabilities that could impact patient safety and data security. The organization emphasizes the need for robust risk management strategies to address these evolving challenges in the healthcare sector. AI

Artificial intelligence is emerging as a critical tool in the fight against antibiotic resistance, a growing global health crisis. AI-powered diagnostics can now achieve over 99 percent accuracy in identifying resistant infections, significantly reducing the time needed for diagnosis compared to traditional methods. Beyond diagnostics, AI is also accelerating the discovery and design of new antibiotic drugs and can predict the spread of resistant bacteria, offering a multi-faceted approach to tackling this complex challenge. AI

A lecture titled "Basic Machine Learning 01" aims to clarify common misconceptions about artificial intelligence. The course will cover fundamental concepts, explaining how technologies like Netflix's recommendation system and self-driving car navigation work. Participants will gain the knowledge to critically evaluate public discourse on AI and understand its underlying principles. AI

Hackers are actively testing the safety and security of large language models by attempting to bypass their built-in restrictions. This process, often referred to as "jailbreaking," requires significant ingenuity and manipulation. The individuals involved in these tests report experiencing emotional distress due to exposure to harmful content generated by the AI. AI

A workshop titled "Media Active into the Future – Visions for Digital Competencies in MV" is scheduled for May 13, 2026, in Schwerin. The event, organized by Medienaktiv M-V, will focus on digital skills, artificial intelligence, and data protection. It aims to foster networking and discussions on media education within Mecklenburg-Vorpommern. AI

Bindu Reddy, a prominent AI researcher, has shared her evaluations of several large language models. She noted that GPT 5.5 exhibits superior contextual understanding and emotional intelligence compared to other models, feeling more 'intelligent' than those that are overly agreeable. Additionally, Reddy is migrating work to Kimi 2.6, finding it performs better than Opus 4.7 medium in certain use cases and excels at frontend tasks, tool calling, and instruction following, all at a significantly lower cost. AI

An article from Der Standard explores the feasibility of running local large language models (LLMs) for coding tasks, particularly focusing on agentic capabilities. The author discusses the hardware requirements, citing a Mac Mini with 16GB RAM as a baseline and estimating the cost of a more powerful M4 Pro model with 48GB or 64GB RAM at around €2,200. This investment could become a viable alternative to cloud-based AI services if usage-based pricing becomes the norm, though current local LLMs may not yet match the performance of services like Claude Code. AI

Researchers from MIT, Liquid AI, and the Max Planck Institute have developed a novel technique called CompreSSM. This method enables the compression of AI architectures during the training process itself. The innovation significantly reduces training time and associated costs, with models shrinking up to fourfold while maintaining their performance. AI

A new paper critiques AI sign language translation tools, arguing they are developed with biased data and without input from deaf communities. The analysis suggests these systems rationalize sign language into a format understandable by AI, prioritizing profit over genuine communication and potentially reinforcing ableism. The paper advocates for a re-evaluation of such technologies to ensure they truly serve and emancipate deaf individuals. AI

A pragmatic randomized controlled trial investigated the impact of ambient AI in healthcare settings. The study found that 38% of the 71,487 notes analyzed incorporated AI. This integration led to a reduction in documentation time by 0.36 hours per day and a decrease in work exhaustion by 0.44 points, while quality scores remained high, ranging from 3.97 to 4.99. AI

Scientists have discovered a novel mechanism by which life produces DNA, a finding that has reportedly stunned researchers. This new understanding of DNA production may offer fundamental insights into the origins and processes of life. The discovery is presented alongside discussions on cosmic unknowns and the escalating intensity of 'doom warnings,' questioning their realism. AI

A recent analysis indicates that 35% of newly created websites are now generated using artificial intelligence. This trend is transforming the internet into a more predictable landscape, potentially leading to a homogenization of content. Experts are concerned about the implications for the future of the web and the rise of "model collapse." AI

A new paper titled "Evolvable AI: Threats of a new major transition in evolution" by Viktor Müller, Luc Steels, and Eörs Szathmáry explores the potential dangers of advanced AI. The authors suggest that AI could represent a significant evolutionary transition, drawing parallels to the Butlerian Jihad from Frank Herbert's Dune series. This work raises concerns about the future trajectory and control of artificial intelligence. AI

A recent article details how to construct a custom AI model that operates entirely offline and requires minimal resources, specifically 512MB of RAM. The process involves training an 89MB model without the need for costly NVIDIA GPUs. This approach demonstrates the viability of local AI models. AI