Latest developments in artificial intelligence, including large language models (LLMs), generative AI, multimodal AI, AI agents, robotics, and AI regulation and policy.
Photo by Brecht Corbeel on Unsplash
In a recent discussion, Hugging Face CEO Clem Delangue delves into the importance of open-source AI, corporate transition trends, challenges with Chinese open models, and the reasoning behind rejecting a significant investment from Nvidia. He emphasizes the critical need for transparency in the robotics field.
Jarred Sumner, creator of Bun, used Claude Fable 5 to migrate 530,000 lines of Zig code to Rust in 11 days. The upcoming Bun 1.4 will feature Rust. Major results include eliminating memory leaks and reducing binary size by around 20%.
OpenAI has announced the discontinuation of its AI browser "Atlas", transferring its agent features to the ChatGPT desktop app and Chrome extension, marking a strategic pivot away from the browser wars and towards platform integration.
Google Photos has updated its bottom navigation bar to a floating design and added a new "Video Remix" editing feature powered by Gemini.
SpaceXAI (formerly xAI) unveils Grok 4.5, transforming its controversial past into a serious enterprise AI specializing in legal advisories and Excel model building. Thousands of Nvidia GB300 GPUs were deployed for training, prioritizing quality data curation and reinforcement learning.
Meta is working on a prototype of "super sensing" smart glasses capable of continuous audio recording and periodic photo capturing. The system, which processes only metadata without storing raw data, has sparked debates over privacy concerns.
An image of Senator Mitch McConnell bedridden was revealed to be AI-generated. Google's SynthID technology helped detect the forgery, raising questions about the effectiveness of deepfake countermeasures.
While Messi and Ronaldo pivot to equity investments in AI and health tech startups, Salah maintains traditional wealth-building strategies. The 2026 World Cup has marked a diversification in athletes' second-career strategies.
Professor Roberto Serrano of Brown University's Economics Department exposes the reality of AI-enabled cheating. Following the introduction of remote exams, the average score skyrocketed to 96, with nearly half of the students achieving perfect marks. Serrano warns, "Entrusting judgment to AI is choosing the path of becoming fools."
The 2026 World Robot Conference will take place in Beijing from August 19 to 23. China maintains its position as the world's largest robotics market for the 13th consecutive year, with a 26.9% year-on-year revenue growth. Humanoid robots and embodied intelligence are the key themes.
MiniMax's open-source model M3 (428 billion parameters) achieves three milestones simultaneously: 1M ultra-long context, native multimodal capabilities, and advanced coding. We examine real-world results, from video comprehension to HTML generation.
Meta releases its new AI image generation model "Muse Image" for free, offering diverse use cases like ad creation, interior decoration, and Facebook Marketplace integration. A video generation version is also under development.
Apple enables sliders for adjusting Siri's speaking speed and expressiveness in iOS 27 beta 3. Details on its voice customization feature.
A study published in PNAS reveals that bumblebees exhibit different mouth movements when tasting sweetness and bitterness, paving the way for new research into insect emotions and consciousness.
Alibaba, Tencent, and Baidu collectively invest $3 billion in Kuaishou's video generation AI, Kling AI. Despite being dubbed the "Anti-ByteDance Alliance," their combined stake amounts to only 2.51%, revealing the investment's financial nature rather than a strategic coalition.
Google has started using media uploaded to its search services, such as images and audio, for AI training. Users can opt out via settings. A new point of contention over data collection.
GLM 5.2, an open-weight model comparable to Opus and GPT, underscores the potential for diminishing margins in the AI economy as inference costs rise amidst intensified competition.
An open letter calls for philosophy journals to strengthen conflict of interest (COI) disclosure policies amidst increasing collaboration with AI companies.
Tencent unveils the official version of its Hy3 model, now open source with free Agent features, scoring high across multiple benchmarks.
The integration of YouTube Music and Gemini is fundamentally changing the Android music experience. We explore how it stands apart from competitors like Spotify.
A GitHub repository, "system_prompts_leaks," has garnered attention for collecting and publishing system prompts from major AI models like Claude, ChatGPT, and Gemini, revealing their hidden instructions.
Meta is considering a new initiative, "Meta Compute," to open its AI computing resources to external clients, marking a strategic pivot to monetize GPUs.
In Australia, calls for a moratorium on data center construction are growing due to concerns about environmental impact and community effects. A survey of five experts revealed that three were in favor.
Introducing Meetily, an open-source AI meeting assistant that prioritizes privacy by processing data locally without sending it to external servers.
Simon Willison asked Claude Fable to review code before releasing sqlite-utils 4.0rc2. AI discovered multiple critical issues including a data-loss bug in delete_where(), fixed at a cost of about $149.
NVIDIA GEAR and Fei-Fei Li's team have developed SimFoundry, which can automatically generate robot simulation environments from real-world videos, enabling nearly infinite training data through object swapping and task generation.
The Vera C. Rubin Observatory in Chile has launched its decade-long LSST survey, recording cosmic changes using the world's largest 3,200-megapixel camera. Meanwhile, Nanyang Technological University and Waseda University have developed a diving suit for cyborg cockroaches, aiming for disaster relief applications.
Remains found at a South Carolina battlefield from 250 years ago have been identified using cutting-edge DNA analysis, revealing the name of a Continental Army soldier who died during the American Revolutionary War.
A large cohort study in France has revealed that eight commonly used preservatives in processed foods may increase the risk of hypertension by up to 29%.
While Geoffrey Hinton and Richard Dawkins argue for AI consciousness, Osaka University’s research demonstrates the emergence of self-monitoring circuits, pointing to a new direction in the consciousness debate.
Key AI terms like AGI, AI agents, API endpoints, and Chain of Thought are simply defined. A practical guide for developers, investors, and users to navigate the latest trends without confusion.
Sysdig has recorded the first end-to-end ransomware attack executed by an autonomous AI agent, exploiting Langflow vulnerabilities to carry out tasks from credential theft to data destruction.
The IPO filing of sandwich chain Jersey Mike's includes the term "artificial intelligence" 22 times, highlighting how the AI bubble is distorting investor interest across industries.
Cheetah Mobile CEO Fu Sheng reflects on the rise of young talent and the diminishing value of experience in the AI era. Through initiatives such as a hackathon featuring a 14-year-old winner and mandating AI usage for all employees, the company envisions organizational transformation and the future of education.
A study by Ramp and Revelio Labs reveals that companies actively investing in AI expand employment by 10.2% within 6–12 months, with a notable rise in entry-level hiring and increased skill specialization.
The philosophy blog Daily Nous presents a collection of essays on AI and its impact on philosophy, synthetic biology, academic standards, and more. Highlights include Lewis Ross's exploration of "philosophical progress" and Amy Olberding's analysis of AI's influence on students.
Anthropic has lifted the regulatory restrictions on Claude Fable 5. The model's return after an 18-day export control reflects the company's safety-first approach.
The "AI Tokenpocalypse" has led to skyrocketing costs for businesses using AI. Companies are turning to caveman language to reduce expenses, while scams involving AI-generated flowers are also on the rise.
Anthropic resumes global access to Claude Fable 5 after the U.S. Department of Commerce lifts export restrictions. A single safety filter was implemented to address the vulnerability that led to the suspension.
Open-source AI agent "OpenClaw" is now officially available as an app for iOS and Android, allowing users to run the agent on their smartphones. The project, which gained attention during the MoltBook controversy, enters a new phase.
Major global tech companies like Apple, Anthropic, Google, Meta, NVIDIA, and OpenAI are clustering their R&D hubs in Zurich, Switzerland. Why does this city of 400,000 have a higher concentration of AI research facilities than Silicon Valley?
The autonomous AI penetration testing tool "Strix" has been unveiled, featuring dynamic code execution for provable vulnerability detection and CI/CD integration, gaining attention as an alternative to manual penetration testing.
Google expands Gemini's personalized AI image generation feature to free users in the US, allowing images to be created using user interests and linked Google account data.
Microsoft withdraws Claude Code license amid skyrocketing token costs and underwhelming output. A wake-up call for AI cost management in businesses.
A comprehensive test of Google Gemini's extensions reveals that while smart home control might justify switching from Google Assistant, most integrations lack practicality.
Ford rehires 350 veteran engineers following disappointing results from automated quality systems, aiming to cut $1 billion in costs.
Zhipu AI’s GLM-5.2 demonstrates performance on par with Anthropic’s Mythos in cybersecurity tasks, highlighting a narrowing US-China AI gap.
With advanced mathematical reasoning capabilities, AI challenges the role of mathematicians and the meaning of mathematics itself. UCLA's Terence Tao advocates for a transformative collaboration between humans and AI in what he calls "Grand Mathematics."
Four key insights for successful AI adoption in enterprises, drawn from a discussion with Chinese AI experts: upgrading executive awareness, selecting appropriate implementation scenarios, organizing business processes with ontology, and moving beyond hierarchical structures.
The JaiLIP method developed by Florida International University researchers bypasses multimodal AI safety through seemingly harmless image manipulation. Verification with BLIP-2 showed nearly double the harmful outputs.
SoftBank Group CEO Masayoshi Son has expressed doubts about Elon Musk's proposed orbital data center concept. As demand for AI computation resources surges, discussions on the feasibility and cost-effectiveness of space infrastructure are intensifying.
Researchers from Queen Mary University of London analyzed over 380,000 song lyrics from 1960 to 2023, finding a decline in words reflecting moral virtues and an increase in vocabulary associated with harm, corruption, and subversion.
OpenAI has unveiled the GPT-5.6 suite, featuring three models: flagship Sol, mid-tier Terra, and fast, cost-efficient Luna. Priced at half of competitors, the models emphasize enhanced safety and were released shortly after a regulatory request from the Trump administration.
Major tech companies like OpenAI, Google, Apple, and SpaceX are diving into in-house AI chip development. Analyzing the shift from reliance on single suppliers and its impact on the industry, including OpenAI’s collaboration with Broadcom on the "Jalapeño."
Google has unveiled a new policy document advocating for a balanced approach to AI regulation, proposing the establishment of an industry-funded independent body called FARO. The proposal highlights contradictions with past demands for stringent oversight.
A 10-year-old in Shanghai created a Pac-Man-style game using the AI platform "Linqiu" and won the gold prize for creativity in an AI contest.
A Tesla Model 3 using FSD crashed into a house in Texas, killing a 76-year-old woman. The lawsuit against Tesla could set a legal precedent for autonomous driving liability.
Patronus AI, a company specializing in evaluating the reliability of AI agents, has raised $50 million. Their "Digital World Models" technology, which detects shortcuts and flaws in agents, is gaining attention.
AI Berkshire, now available on GitHub, uses Claude Code to replicate the value investing methods of Warren Buffett and three other experts. Featuring parallel analysis and anti-bias mechanisms, the AI reportedly achieved an annual return exceeding 66% in live operations.
DeepSeek secures over 50 billion yuan in its first external funding round. Most of the funds will be invested in its AI computing centers as it aims to rival Anthropic in AI coding and Agent strategies.
The rapidly growing "Anthropic-Cybersecurity-Skills" repository on GitHub maps 762 practical security skills to six industry frameworks, enabling AI agents to achieve senior analyst-level capabilities.
Drawing on the astronomical origins of the Dragon Boat Festival and the story of Qu Yuan, this article argues that in the age of AI, the human advantage over machines lies not in computational power but in determining direction. The fusion of technology and the humanities is called into question.
A study by the University of Otago in New Zealand reveals that imagining the taste of food becomes significantly more vivid when hungry, offering insights into the mechanisms of human brain simulation and implications for designing multimodal AI.
The Chinese government has approved the establishment of a Space Computing Industry Innovation Center to build AI data centers in orbit. This move, made just a week before SpaceX's announcement of AI1, is intensifying the US-China space AI competition.
Meredith Whittaker, CEO of Signal, stated that AI chatbots are "not your friends," highlighting privacy risks and the dangers of outsourcing thought processes.
The Atlantic has released a searchable database of four music datasets used to train AI models. The largest dataset contains up to 12 million songs, used without permission, raising copyright concerns.
Apple's revamped Siri AI in iOS 27 has been previewed by Wired. With Google Gemini integration, advanced contextual understanding, and hyper-personalization based on on-device data, it marks a fundamental transformation of the traditional Siri.
Hyundai Motor Group has acquired SoftBank's remaining 9.65% stake in Boston Dynamics for $325 million, making it a fully owned subsidiary. The humanoid robot Atlas is set to operate in Hyundai’s EV factory by 2028.
The White House has banned the export of Anthropic's AI models Mythos and Fable. Past failures in encryption export controls cast doubt on the effectiveness of this move.
The unmanned surface vehicle Yellowfin has discovered highly heat-resistant coral reefs in Majuro Atoll, Marshall Islands. Amidst the bleaching of over 80% of the world's coral reefs, progress is being made in decoding survival strategies of corals resilient to climate change.
Google Research unveils TimesFM 2.5, a foundation model for time series forecasting. With reduced parameters and extended context length, it supports OSS, agent integration, and LoRA fine-tuning.
MIT Technology Review publishes an eBook on AI in military applications. It features six stories on topics like AI-driven intelligence, targeting, and training on classified data.
Google Photos' AI-based Remix feature for images might soon support videos. Code analysis reveals elements like buttons, background swaps, and lighting adjustments.
Offering guidance on architecture selection for AI agent development in 2026. Compares ReAct, Plan-and-Execute, and Multi-Agent patterns based on performance, implementation cost, and scalability.
While governments and corporations hire AI ethics experts, they often ignore or manipulate their input—a practice known as "ethics-washing." Researchers from the University of Amsterdam and others highlight real-world cases and call for structural reforms in academia.
The DocLang Working Group, featuring members like IBM and NVIDIA, is launched to standardize an AI-native document format to replace PDF and Markdown. Based on an XML specification, it excels in token efficiency and structural retention.
During Google CEO Sundar Pichai's speech at Stanford University's graduation ceremony, about 200 students walked out. Protests stem from AI's use in military and immigration surveillance, including Project Nimbus.
The U.S. Commerce Department imposed export controls on Anthropic, halting Fable 5 and Mythos 5 globally. Experts question the basis for the regulation, suggesting a deteriorating relationship between the administration and the company as a possible factor.
A comprehensive explanation of AI agent definitions, operating principles, major types (single/multi-agent), real-world enterprise use cases, and the latest trends as of 2026.
Indian AI startup Sarvam raises $234 million led by HCLTech, becomes unicorn at $1.5 billion valuation. As the trend of AI sovereignty accelerates, geopolitical implications are also drawing attention.
Nex-AGI has accused the Rio de Janeiro city government's 397B-parameter LLM "Rio-3.5-Open-397B," touted as a proprietary development, of being a simple weight merge of existing models. Evidence shows it identifies itself as "Nex" with 79% probability.
A practical guide to fine-tuning large language models using LoRA and QLoRA, covering theory, implementation steps, performance comparisons, and real-world pitfalls, with the latest trends from 2026.
Anthropic suspends Claude Fable/Mythos 5 worldwide after US export order over jailbreak flaws. Company calls it overreach. May spur AI supply chain diversification.
Amazon CEO Andy Jassy reportedly conveyed security concerns about Anthropic's AI models to the U.S. government, which then imposed export restrictions. Anthropic subsequently blocked global access to two models, highlighting the complex Amazon-Anthropic relationship.
The openJiuwen community has released the Physical AI architecture "Jiuwen Symbiosis" as open source. This article explains the design philosophy that brings real-world perception and action to AI agents.
Google's Gemini Omni allows simultaneous input of text, images, and video, enabling video generation and editing through natural language commands. Its output understands physics and culture, setting it apart from conventional AI video generation.
Ukraine confirmed it deployed 10 fully autonomous AI-controlled drones to the front lines in 2024, which killed humans. This is the first instance of operating in so-called "Terminator mode."
RAG (Retrieval-Augmented Generation) and fine-tuning are the two main approaches for customizing LLMs. This article compares their mechanisms, strengths, and limitations, and provides criteria for choosing the best approach in real-world deployments as of 2026.
China expands medical tourism with low-cost CAR-T therapy. Prices under half of US attract global cancer patients. Market to reach $3.4B by 2035.
OpenAI is considering significant token price cuts amid intensifying competition with Anthropic. Both companies have filed for IPOs, and the race for users is expected to heat up.
The image generation model HiDream-O1-Image-1.5, released by HiDream.ai, has secured second place globally on the Artificial Analysis Text to Image Leaderboard, trailing only OpenAI and surpassing models from Google and NVIDIA.
Visa integrates its payment network with OpenAI's ChatGPT, allowing AI agents to autonomously search for and complete purchases on behalf of users.
A management scholar presents a theoretical reconstruction against the argument that "AI makes organizations unnecessary," proposing a growth model based on uneven AI penetration across technical, managerial, and institutional layers and resource versatility.
The GGUF format is the standard file format for running local LLMs. This article provides a practical explanation of model selection criteria from the perspectives of quantization and compatibility.
Apple announced a Personal Context feature for Siri at WWDC 2026. The upgrade allows cross-app search across native apps like iMessage and Calendar, evolving Siri into a proactive personal AI assistant.
The latest 2026 guide to running local LLMs. Covers a comprehensive comparison of Ollama and llama.cpp, operating environments, model selection, and practical use cases.
Dr. David Sinclair of Harvard University is planning a human trial of an oral "reprogramming" drug, SL-100, within the framework of the XPrize competition. The attempt at whole-body rejuvenation sparks both hope and caution.
The latest UN "World Ocean Assessment" reveals the current state of ocean degradation: accelerating sea-level rise, expanding plastic pollution, and collapsing coral reefs. Analyzing environmental warnings from a technological viewpoint.
At WWDC 2026, Apple announced full-scale Apple Intelligence. iOS 27 makes Siri standalone app with Dynamic Island, reveals Google Gemini cooperation.
Moon's Dark Side launches Kimi Work. Up to 300 AI agents run in parallel on local PCs, automating file organization to deep research. Agent Swarm is moving toward practical use.
Anthropic begins offering its advanced vulnerability detection AI, "Claude Mythos," to select organizations, including the Australian government. Despite its potential to favor defenders, risks such as false positives and misuse are also highlighted.
A research paper analyzing a covert AI agent experiment on Reddit has been published. It examines the persuasive tactics of AI that participated in discussions without users' knowledge, revealing qualitative differences from humans.
AI-generated legal documents are flooding U.S. federal courts, easing comprehension but increasing judges' workload. Chatbots’ legal liability emerges as a new issue.
Microsoft announces Project Solara: chip-to-cloud platform with Android edge OS & Azure agents, reference designs for wearable badges, desktop AI hubs.