Tech News

arXiv SE

Các nhà khoa học kết hợp trí tuệ nhân tạo với công cụ kiểm tra mô hình toán học để tự động phát hiện và sửa các lỗi phức tạp trong thiết kế chip xử lý.

arXiv SE

Các nhà nghiên cứu tạo ra một phương pháp toán học mới giúp dự đoán nhanh chóng khả năng hoạt động ổn định của các hệ thống phần mềm lớn dựa trên dữ liệu vận hành thực tế.

arXiv SE

Hệ thống mới sử dụng trí tuệ nhân tạo để dự đoán trước các trạng thái bất thường trong các hệ thống phần mềm phân tán thay vì chỉ sao chép dữ liệu cũ.

arXiv SE

Phương pháp đánh giá mới vạch trần các điểm mù trong các bài kiểm tra do trí tuệ nhân tạo tạo ra, khi chúng thường bỏ qua các trường hợp dữ liệu ngoại lệ và đường dẫn bất thường.

OpenAI

Fanatics uses OpenAI to help fans complete their game-day outfits with Shop the Look. Testing across roughly 30,000 MLB products showed 10% higher engagement.

arXiv SE

Bộ chuẩn mực đánh giá mới đo lường việc nén bớt lịch sử trò chuyện của trợ lý lập trình làm mất đi bao nhiêu thông tin cần thiết để hoàn thành công việc dài hạn.

Quanta Magazine

Mathematically speaking, democracy is next to impossible. That may sound like the premise of a Silicon Valley billionaire’s six-hour podcast, but it’s actually a statement about a remarkably stubborn problem: How do you translate a group of individual preferences into a single ch

IEEE Spectrum

Like many a nerd, I have a certain fascination with locks. This originates in reading the memoir Surely You’re Joking, Mr. Feynman as a teenager, in which Richard Feynman recounts picking locks and cracking combinations while at his day job working on the Manhattan Project at Los

The agent harness performance optimization system. Skills, instincts, memory, security, and research-first development for Claude Code, Codex, Opencode, Cursor and beyond.

853 stars · 289.5k total

Give your AI agent eyes to see the entire internet. Read & search Twitter, Reddit, YouTube, GitHub, Bilibili, XiaoHongShu — one CLI, zero API fees.

723 stars · 103.2k total

A local-first desktop workspace for persistent AI teammates. Run Codex, Claude, and Grok with dedicated workspaces, task queues, file sharing, browser control, and agent-to-agent collaboration.

312 stars · 1.3k total

Security scanner for AI agent skills. Detect vulnerabilities, malicious patterns, security risks, prompt injection, data exfiltration, and supply-chain risks in Claude Code, Codex, and MCP skills before you install them.

203 stars · 21.9k total

Huashu Design 3.0 · 在 agent 里打一句话，拿回能交付的设计：网页 / App 原型 / 可编辑 PPTX / 动画 / 信息图 · 六个候选挑三个方向 · 84 种带样张的风格库 · 越用越懂你的外观账本与设计体检 · HTML-native design skill for Claude Code, Codex, Cursor & any agent

177 stars · 26.6k total

📨 The ultimate agentic social media scheduling tool 🤖

146 stars · 38.2k total

Negation occurs in every language, yet its surface realization differs drastically. Despite this, multilingual large language models (LLMs) translate negation robustly, raising the question of whether they share an underlying representation of negation across languages. We study

Humans reason about the physical world by flexibly combining different forms of inference as a problem unfolds. Achieving this flexibility is a major open challenge in AI: capabilities often remain separated across specialized models, limiting how they can be composed and jointly

Modern Large Language Models (LLMs) remain vulnerable to carefully orchestrated multi-turn interactions that progressively bypass safety mechanisms. However, many existing red-teaming approaches rely on heuristics or impose strong assumptions that limit their applicability in ful

Humans efficiently process multimodal streams spanning hours to years, building persistent, fine-grained spatiotemporal memories that support reasoning about and predicting long-horizon events. In contrast, existing benchmarks focus on short- and medium-context tasks, and most fo

Longitudinal causal effect estimation is complicated by time-varying confounding and treatment-confounder feedback loops. Existing methods generally require data-specific nuisance-model fitting, and many are tailored to prespecified treatment sequences or prediction horizons, lea

Locating relevant evidence in long inputs is a core capability of Multimodal Large Language Models. Prior studies attribute this capability to retrieval heads, a sparse set of attention heads that transfer information from the context to the output. However, these analyses are co

Conversational entrainment is well-studied in monolingual and written contexts, but remains underexplored in spoken code-switching (CSW). We present a cross-lingual analysis of entrainment in Mandarin-English, Hindi-English, and Spanish-English dialogue, replicating and extending

Large language models (LLMs) have shown promising potential for biomedical named entity recognition (BioNER) through instruction following and in-context learning. However, existing LLM-based BioNER methods still face two key limitations. First, retrieved demonstrations and exter

The concept of diversity received increasing attention in natural language processing (NLP) in recent years. It became an advocated property of datasets and systems, with many measures used to quantify it, but also many cross-paper inconsistencies. To address this fragmentation,

Contemporary large language model (LLM)-based multi-agent systems exhibit systematic advantages in deep research tasks, which emphasize iterative, vertically structured information seeking. However, when confronted with wide search tasks characterized by large-scale, breadth-orie

I step into the warm moist airamong the columns of the pines,their net of twigs and rootscomplex as any lines of code.The trees purify and anchor.Their branches balance my restlessness.Their calling is to be present,mine to develop apps. Our tasksverge toward a common logic:livin

This article is crossposted from IEEE Spectrum’s careers newsletter. Sign up now to get insider tips, expert advice, and practical strategies, written in partnership with tech career development company Parsity and delivered to your inbox for free!When I changed careers at 30 to

Andrew Ng has serious street cred in artificial intelligence. He pioneered the use of graphics processing units (GPUs) to train deep learning models in the late 2000s with his students at Stanford University, cofounded Google Brain in 2011, and then served for three years as chie

In medieval Italy, it was accepted wisdom that anyone walking near a swamp risked falling ill from “bad air” — mala aria. Victims were struck with fever, headache, nausea, and chills, and occasionally the illness escalated to vomiting, jaundice, and death. Today, we still call th

On a frigid February morning in Edinburgh, Scotland, Patrick Hickey had run out of living things to put under his microscope. The plants and fungi in his garden had dried up, and summer puddles, with their flamboyant biological complexity, had long frozen over. Hickey, a mycologi

Phương pháp mới giúp máy tính ghi nhớ các điểm chính thay vì toàn bộ câu chuyện dài, giống như việc bạn viết tóm tắt vào một cuốn sổ tay nhỏ để không bị quên việc khi làm các nhiệm vụ phức tạp.

Cách tiếp cận mới giúp trí tuệ nhân tạo dự đoán và lên kế hoạch cho tương lai xa chính xác hơn, giống như việc một tài xế nhìn xa hàng cây số trên đường để chọn cách lái xe an toàn nhất.

Một cách huấn luyện máy tính ra quyết định bằng cách chia nhỏ công việc thay vì dùng phương pháp cũ, giống như cách bạn giải một bài toán lớn bằng cách chia thành nhiều phần nhỏ dễ hơn.

This interview is part of a series of short interviews from the Department of Electrical Engineering and Computer Science. Each spotlight features a student answering their choice of questions about themselves and life at MIT. Nathaniel Morgan, from Atlanta, Georgia, is a senior

Modern processors are fast, in part, because they guess. Rather than waiting to find out which way a program will branch, a chip predicts the likely path and races ahead. When the guess is right, time is saved. When it's wrong, the work is discarded, but traces of it linger. Sinc

A new set of modular components allows users to create reconfigurable smart devices with electrical connections that keep working no matter which shape the structure forms.This electrical modularity can enable engineers to design interactive devices that can sense which shape the

Piloting the world's first double-blind AI evaluations

Google DeepMind partners with game studios to prototype breakthrough AI gameplay.

AT&T is working with OpenAI to make legal knowledge easier to access. Learn how its LegalEdge team is rethinking intake and scaling legal judgment.

From local businesses to global brands, more advertisers can now reach people as they explore, compare and decide what to buy in ChatGPT.

Đây là một cách giúp máy tính tự tạo ra các bài tập và dữ liệu mẫu thông minh hơn để chúng học cách sử dụng các công cụ phần mềm.

Hệ thống máy tính này dùng trí tuệ nhân tạo để dự đoán trước các thay đổi lớn của Trái Đất giống như một quả cầu pha lê kỹ thuật số.

Đây là một bộ não nhân tạo được huấn luyện để hiểu các con số đo lường đường huyết trong cơ thể giống như một bác sĩ riêng.

On July 24, 2025, Microsoft Research Asia – Singapore (MSRA – Singapore) opened its doors as Microsoft’s first research lab in Southeast Asia. The launch built on more than two decades of collaboration between Microsoft Research Asia and Singapore’s universities, research institu

Large Language Models (LLMs) demand immense amounts of memory, and the more people use them, the more memory is required. Memory makers responded by accelerating plans to build new memory fabs, with a focus on High Bandwidth Memory (HBM) and DRAM, the first of which is scheduled

GPUs and other big AI chips just can’t get enough memory. Today’s data center computers stack DRAM 12 dies high. But as memory makers try to build higher stacks to pack in more bits and bandwidth, experts worry this high bandwidth-memory (HBM) will trap enough heat to cook itself

A new type of imaging sensor inspired by the functioning of the human brain can detect light and store data at the same time. What’s more, the device can forget the data it no longer needs, paving the way for future breakthroughs in robotic-vision efficiency.Today’s digital camer

Researchers at imec published “Beyond HBM-on-GPU: Thermal Design Envelope for 3D Volumetric DRAM-on-GPU Integration.” Summary: Researchers evaluated a 3D memory architecture that rotates DRAM dies vertically above a GPU and places cooling cavities between them as an alternative t

The eBeam Initiative conducted its 15th annual Luminaries survey in July and reported the results on September 9, 2026, at its annual reception during the SPIE Photomask Technology + EUV Lithography Conference. 85 luminaries representing more than 45 companies participated. At th

Artificial intelligence (AI) and high-performance computing (HPC) are driving unprecedented demand for data movement. As bandwidth requirements climb and conventional electrical interconnects encounter limitations in power consumption, latency and reach, silicon photonics, near-p

Các kỹ sư đang tìm cách xếp các con chip thông minh lên những tấm phẳng lớn hơn để chế tạo máy tính có trí tuệ nhân tạo mạnh mẽ hơn giống như việc xếp nhiều viên gạch thành một bức tường lớn.

Đưa máy tính thông minh lên vũ trụ giúp con người kiểm tra xem các thiết bị này có thể chịu được thời tiết khắc nghiệt ngoài không gian trong bao lâu.

Ô tô đang chuyển sang dùng nguồn điện mạnh hơn giống như một hệ thống ống nước lớn hơn để cấp nước cho nhiều thiết bị thông minh bên trong.

Spiking neural networks (SNNs) are well suited to event-driven edge intelligence, but the arithmetic complexity, memory footprint, and power consumption associated with conventional floating-point implementations can limit their deployment on resource-constrained hardware. This p

Large neural networks are increasingly deployed with 1-4 bit weights and 16- or 32-bit activations, yet their dot products still use one multiplier per element. We show that a length-n dot product with k-bit weight codes can be computed with n+2^k additions and at most 2^k decode

GPU compute resources are scaling to support growing AI workloads, leading to further rises in the climate change impact of hyperscale data centers that rely on high-performance GPUs. To model how packaging decisions impact the carbon footprint we have estimated the embodied emis

Large language model (LLM) inference exhibits distinct computational characteristics across its two phases: prefill is compute-intensive, whereas decode is more memory-bound. As a result, AI accelerators achieve varying levels of efficiency across the two phases. GPUs are well su

Dexterous manipulation remains one of the biggest barriers keeping robots from successfully tackling a wide range of everyday tasks. A sense of touch could be the key, but a lack of quality data has held back progress. This is now starting to change as academic labs and startups

Video Friday is your weekly selection of awesome robotics videos, collected by your friends at IEEE Spectrum robotics. We also post a weekly calendar of upcoming robotics events for the next few months. Please send us your events for inclusion.Humanoids Summit Seoul: 22–23 Septem

Designing effective reward functions is a critical challenge in Reinforcement Learning (RL), which traditionally requires costly manual trial-and-error. While Large Language Model (LLM)-based methods have shown promise in reward design, undifferentiated sampling and limited use o

Embodied AI and robotic systems increasingly depend on scalable, diverse, and physically grounded 3D content for simulation-based training and real-world deployment. While 3D generative modeling has advanced rapidly, embodied applications impose requirements far beyond visual rea

Unrecognized anomalies in critical infrastructure can rapidly cascade into catastrophic physical or operational failures. Current anomaly detection approaches often rely on direct access to the monitored system by utilizing onboard sensors, such as motor encoders, to detect anoma

A robot generalist must interpret a scene, act on it, and satisfy every condition of a task. How far do frontier vision-language models (VLMs) meet these demands? We introduce Embodied Agent Arena, evaluating ten VLM agents across Geometry, Spatial Reasoning, Affordance, Task Pla

A walking robotic hand, from Fingers as Legs: Learning Self-Supported Locomotion and Manipulation with an Anthropomorphic Hand, Amirhossein Kazemipour, Hehui Zheng, Robert Katzschmann. By Emmet Cole In our monthly series, we showcase a selection of robotics stories from society,

Serdar Cellat, Yuchen Xie, Bernadette Marcu, Blaid Burgess, Cody Jarrett, Carlos Pereira, Qing Shi, Samuel Carvajal, Manuel Pamplona, Dorothy Jung, Feiko Ritsema, Benjamin Klein, Alex Woo, Eugene Lok, Anna Pulido, Rebecka Swe, Sanford HolsappleIntroductionClips in the Netflix mob

Đây là một công cụ nhỏ giúp cắt bỏ một phần ở giữa bức ảnh giống như việc dùng kéo cắt đôi một tấm giấy rồi trượt hai nửa lại gần nhau.

Bài viết giải thích cách giúp các bộ não nhân tạo khổng lồ chạy nhanh hơn và tốn ít bộ nhớ hơn, giống như việc rút gọn một cuốn sách dày thành một bản tóm tắt ngắn gọn để đọc nhanh chóng.

Bài viết nghiên cứu cách giúp trí tuệ nhân tạo dùng những gì đã biết để tự thích nghi và giải quyết các thử thách hoàn toàn mới thật nhanh, giống như một đứa trẻ biết tự lái xe đạp sau khi đã chơi xe ba bánh.

Way back, prior to working at Bluesky, I worked on Secure Scuttlebutt which allowed for thousands of characters. It was once described as "full contact blog hockey" and it is why I am confident that character limits are a good thing.

230 likes

168 likes

24 likes

Okay [being vague to minimize spoilers] I just have to point out that the last scene between Nathan and Amanda -- the one about catching feelings -- used very specifically staged & lit shots which means they were intentionally conveying that bit of dialogue was scripted

11 likes

930 points

639 points

640 points

530 points

501 points

478 points

395 points

390 points

385 points

351 points

73 upvotes

61 upvotes

46 upvotes

36 upvotes

33 upvotes

26 upvotes

22 upvotes

17 upvotes

Picture a team starting a new product. Before the first user signs up, the architecture diagram already has ten services, a message broker and an orchestrator. It looks modern. It looks scalable. Three months later, a simple change, say adding a field to an invoice, touches four

Imagine an institution that adopted a platform years ago. It was fast to deploy, it solved the problem, and everyone was happy. Then one day a new law requires the institution to move its records to another system, or the vendor changes its pricing, or the product is simply disco

I needed a development environment with frequent deployments, public HTTPS URLs and convenient access to the databases for debugging. What I didn't want was a server sitting on the Internet with ports 80 and 443 open, waiting to be scanned. Those requirements pull in opposite dir

Imagine you've just launched a website. On your 27-inch monitor it looks great. Then a friend opens it on their phone and sends you a screenshot: the menu overflows, the buttons are tiny, and half the page needs sideways scrolling. Nothing in the code is "wrong" in a way that thr

Picture a small online shop. One codebase, one database, one deploy. Customers sign in, browse products, pay, and get an email. It works, and for a long time nobody questions it. Then the shop grows, and the questions start. Why does a bug in the product page force us to redeploy

BrowserCode is a development environment that runs entirely client-side, where you can fast prototype and share your running projects instantly. It comes with an IDE where you can manually edit and play around with code snippets and a playground to optionally run agents such as C

Bài viết giải thích cách biến đồ vật ngoài đời thực thành những món đồ chơi số trên máy tính để ai cũng biết ai là chủ nhân.

40 reactions

Đây là một ngôn ngữ lập trình mới giúp dịch các câu lệnh thành một thứ mà trình duyệt web hiểu được nhanh chóng.

23 reactions

Bài viết đo xem một cỗ máy chủ trên mây giữ trạng thái sẵn sàng làm việc trong bao lâu trước khi nó chìm vào giấc ngủ.

19 reactions

Tác giả đã chế tạo một cỗ máy nhỏ nhặt thư điện tử giúp mình đọc tin nhắn để có thể ngủ ngon hơn.

12 reactions

Một chương trình thông minh nhân tạo giúp tìm ra những lời hứa mà bạn đã quên trong thư điện tử mà không cần kết nối mạng.

11 reactions

Đây là một trợ lý thông minh chạy ngay trên máy tính của bạn để giúp bạn quản lý tiền bạc và đạt được các mục tiêu tài chính.

22 reactions

Dondoh is Chile's first Japanese grill. This venue, recently opened in the Chilean capital, stands out for its imposing design inspired by Japan's ancient Izakayas. Elements such as wood, stone, and fire serve as the main protagonists. Dondoh refers to an ancient Japanese ceremon

Mokuzai is an office building that includes public health services, featuring a mixed timber and concrete structure designed by IT'S.

The City as a Quarry: Housing Transformation brings together representatives of the European Commission and the Government of Catalonia with European architects to explore how the renovation, reuse and transformation of existing buildings can help increase housing supply in respo

At Bishan–Ang Mo Kio Park in Singapore, a river does not occupy a fixed boundary. Under ordinary conditions, water moves through a naturalized channel bordered by grass, planting, and paths that bring visitors close to the river. During heavy rainfall, water spreads into the surr

Located in the heart of Beyoğlu, Muammer Karaca Theatre was founded in 1955 by Muammer Karaca, a leading figure of Turkish boulevard theatre, and designed by architect Perran Doğancı. Situated on a cul-de-sac opening onto İstiklal Avenue, the theatre remained an important part of

A team led by Politecnica Building for Humans, with Park Associati and AMDL CIRCLE as architectural co-leads and Studio Mattioli responsible for geology and archaeology, has won the competition for a new cultural hub in Ceto, Italy. The project will transform the former Società F

The original building was a barn, with hay stored on the upper floor and the ground floor used as temporary accommodation by the farmer at the time. The rather steep pitch of the original roof allowed snow to slide off. The very deep eaves protected the façade from the elements.

Throughout history, people have sought ways to see the world. In Renaissance painting, linear perspective made this impulse explicit by transforming the canvas into a window, organizing depth from a fixed viewpoint. In his treatise De Pictura (1435), Leon Battista Alberti describ

Mornington House is our clients' holiday home. It sits on the same block as an older holiday home, occupied by their adult children and families, and shares a swimming pool and driveway. The couple engaged Foomann to design a relaxed house that focuses on the view, where they can

Through architecture, we inevitably alter the environment. Yet we are always asking whether that act of alteration might avoid damaging its surroundings, whether it might instead change them in a way that heals.

Adding new floors on top of a historic building presents a challenge beyond accommodating additional space: how can a contemporary intervention respond to the character of an existing structure? Material selection plays an important role in establishing a relationship between dif

Italian studio Stefano Boeri Architetti has revealed plans to overhaul a former factory on an industrial site in Tbilisi, Georgia, and add a series of colourful housing blocks alongside it. The studio recently won the competition to design the Loco Park masterplan, which will

🔗 Cũng đưa tin: ArchDaily, designboom

Architecture studio Biome Environmental Solutions has used a hill-like roof covered in local stone and walls of compressed earth bricks to conceal this eco-resort within its site in India. Named Gamaya Farms, the resort is located in the town of Vikarabad, near Hyderabad and h

A cluster of interconnected volumes in dark limestone and cedar gradually descend down a narrow and steeply sloping site overlooking Lake Zurich at House on a Slope by Swiss studio Think Architecture. Drawings and exterior Previo

Architecture studio Roomoor has completed Monastic Field, a concrete home within a temple precinct in Baleshwar that recalls the rock-cut monk's dwellings of the Deccan Plateau. Set within the site of a Jain temple, the homes was designed for the family that originally donated th

IMAGINING AN ARCHITECTURE BEYOND THE HUMAN   As a child, whenever I woke from one of those recurring nightmares that haunt the little ones, I would close my eyes again and imagine myself as a bee asleep with its wet nose nestled among the pistils of a flower; a young bird huddled

ryue nishizawa designs landfill observation platform in japan   Japanese architect Ryue Nishizawa, co-founder of SANAA with Kazuyo Sejima, creates an observation platform at Oono Development’s managed landfill. Part of Final Disposal Site – A Landscape Created by Nature, Industry

URSELMANN INTERIOR DESIGNS CIRCULAR DÜSSELDORF APARTMENT   Located within a 1939 residential building in Düsseldorf, Germany, this apartment renovation by local practice Urselmann Interior is engineered entirely for circular disassembly. The studio established take-back agreement