Tech News

Simon Willison

Báo cáo thường niên của GitHub cho thấy nhờ sự trợ giúp của trí tuệ nhân tạo, số lượng chương trình máy tính và đoạn mã do lập trình viên tạo ra đã tăng vọt chưa từng có trong năm qua.

OpenAI

See how Legora uses OpenAI models, ChatGPT, and Codex to accelerate legal workflows, expand contract analysis, and free lawyers to focus on higher-value work.

🔗 Cũng đưa tin: OpenAI

Google Research

Các nhà khoa học tạo ra một cách giúp trí tuệ nhân tạo vừa học điều mới mà không bị quên đi những kiến thức cũ đã học giống như cách con người ngủ để sắp xếp lại trí nhớ.

Simon Willison

Các mô hình trí tuệ nhân tạo đời mới đã thông minh hơn rất nhiều, giúp mọi người có thể dễ dàng tạo ra những trò chơi điện tử vui nhộn mà không cần phải biết cách viết mã.

Hugging Face

IEEE Spectrum

In the cockpit of Beta Technologies’ Alia CX300, pilot Chris Caputo steers onto Runway 33 at Patrick Leahy Burlington International Airport in Vermont. Cleared for takeoff, he releases the brakes and nudges the throttle forward.At the back of the aircraft, the five-blade pusher p

Lil'Log (Lilian Weng)

Bài viết trình bày phương pháp dạy robot tập luyện trong thế giới ảo trước, sau đó thay đổi liên tục các điều kiện giả lập để nó có thể tự tin bước ra thế giới thực.

Simon Willison

Các kỹ sư đã sử dụng trí tuệ nhân tạo để viết chương trình giúp một chiếc thẻ đeo hội nghị nhỏ xíu có thể chạy được trò chơi điện tử máy cầm tay cổ điển.

Công cụ này dùng sóng wifi trong nhà giống như một tấm lưới tàng hình để nhận biết có người đang đứng ở đâu và thở ra sao mà không cần dùng đến máy quay phim.

596 stars · 100.8k total

Bộ tài liệu này giúp các trợ lý thông minh nhân tạo hiểu rõ cách tạo ra các ứng dụng điện thoại để viết mã lệnh chuẩn xác hơn.

212 stars · 3.4k total

Prostate cancer is among the most frequently diagnosed malignancies in men worldwide. However, precise prediction of biochemical recurrence (BCR) after radical prostatectomy remains challenging. Here, we propose a computationally efficient multi section whole slide image analysis

Distribution-matching distillation enables high-quality image generation in only a few denoising steps. However, distilled student models often map different initial noise samples to similar outputs under the same prompt. To locate where diversity is lost, we compare the teacher

Do vision-language models (VLMs) share our ability to communicate visually? We share preliminary findings, starting from a surprising observation: simple visual symbols like a red heart can strongly influence model behavior on seemingly unrelated tasks, such as steering models to

Multi-Agent Path Planning (MAPP) requires collision-free paths for multiple agents. MAPP has numerous real-life applications, including the navigation of automated vehicles, and has therefore many variants considering different constraints, requirements, and objectives. For each

Skin cancer is among the most prevalent and life threatening forms of cancer worldwide. Early and accurate diagnosis significantly improves patient survival rates. However, manual dermoscopic assessment remains challenging due to visual similarity. This study proposes an Explaina

We release a battery of 28 propensities for AI research, with expert-written rubrics, synthetic datasets, and annotated demand intervals, built with the method of Romero-Alvarado et al. (2026). We selected a wide collection of generic propensities that may influence the behavior

Deep generative models reproduce the observational distribution underlying the training data, which, in turn, inherit any spurious associations it contains. A common source is an unobserved confounder that simultaneously affects both a treatment attribute the user wants to contro

Machine learning systems trained on untrusted data are vulnerable to data and backdoor poisoning attacks, in which an adversary corrupts training and test samples to manipulate predictions at inference time. Randomized smoothing yields provable defenses against such attacks, but

Medical AI has demonstrated specialist-level diagnostic accuracy, yet these capabilities remain largely inaccessible in resource-constrained rural settings where bandwidth is scarce, compute is limited, and clinical decision-making requires integrating heterogeneous modalities. W

Feature analysis is a crucial step in medical image analysis, particularly in breast cancer classification. Traditional image analysis methods often struggle with optimal patterns along the diagonal axis, computational efficiency and feature buffer storage. Different classical op

We measure, without sampling error, the probability a language model assigns to each integer from 1 to N when asked for a random number. For a fixed prompt the model defines one probability vector, which we compute by summing over each integer's scored spellings rather than estim

When expanding a report, writing with agents may introduce a stronger claim than the article makes and then disclaim it. This is called over-defensive writing. We assume that this is because the agent is writing for responding to an adversarial AI reviewer, instead of a human rea

The Asch conformity experiment has recently gained considerable attention as a way of measuring conformity in large language models (LLMs). While previous studies have focused on demonstrating conformity in LLMs, our work adds a further quantitative layer to this approach. Instea

The availability of a wide range of large language models (LLMs) embedded in various agentic systems has significantly increased the potential of model selection strategies to improve the cost-performance tradeoff. Existing strategies involve either routing, where a single model

Cross-modal translation requires models to capture shared semantics across modalities while preserving the fine-grained structure required by the target modality. Yet in real-world scientific domains, paired data are often imperfect, while obtaining clean pairs at scale is costly

Open-source vulnerability mitigation spans upstream libraries and the downstream projects that depend on them. Upstream, a vulnerability progresses through a public report when one exists, a fix, a fixed release, and NVD disclosure. Downstream, a project mitigates it by adopting

Over the past decade, the significant advancements in artificial intelligence have been closely tied to increasingly sophisticated training frameworks and steadily powerful computing hardware. However, the entire ecosystem remains fragmented. On one hand, models are represented w

Large language models (LLMs) are increasingly used for code generation, motivating techniques that guarantee the structural and semantic correctness of generated programs. Constrained decoding enforces correctness during generation by restricting token choices to those that admit

AI coding agents such as Claude Code, GitHub Copilot coding agent, and Cursor now commit directly to open-source repositories. Empirical studies of these agents identify their commits through authorship signatures, such as bot accounts and Co-authored-by trailers. These signature

Agentic coding tools, such as OpenAI Codex, Claude Code, and Cursor, are changing how software is developed. These AI-powered systems function as autonomous teammates capable of planning and executing complex development tasks. Agents have become active participants in refactorin

If you’ve ever wondered how Claude and similar AI agents can answer complex questions in seconds and where they get the information, you’re not alone.Gurmeet Saran works to find that answer every day as an engineer at Anthropic in San Francisco, the AI research company behind Cla

This article was originally published by Canary Media.The geothermal startup Fervo Energy is now selling power to the grid from its Cape Station project in Utah—marking the first time a utility-scale, next-generation geothermal facility has achieved such a feat worldwide. The Hou

This White Paper provides roboticists and control engineers with a comprehensive overview of how motion planning and motion tracking can be decoupled to preserve professional athletic style in humanoid robots, and why adaptive, speed-aware control is essential for closing the gap

Congratulations to the Berkeley Artificial Intelligence Research (BAIR) Lab class of 2026! This year, BAIR celebrates another remarkable group of Ph.D. graduates whose curiosity, creativity, and perseverance have pushed the frontiers of artificial intelligence and machine learnin

Overview of adaptive parallel reasoning. What if a reasoning model could decide for itself when to decompose and parallelize independent subtasks, how many concurrent threads to spawn, and how to coordinate them based on the problem at hand? We provide a detailed analysis of re

Understanding the behavior of complex machine learning systems, particularly Large Language Models (LLMs), is a critical challenge in modern artificial intelligence. Interpretability research aims to make the decision-making process more transparent to model builders and impacted

RNA vaccines, which have been proven effective against Covid-19, are now being developed for many other diseases, including cancer. One of the drawbacks to these vaccines is that they require ultracold storage, but researchers from MIT have found a promising way to overcome that

Over the past 80 years, America’s bold, sustained investment in scientific research, and the discoveries, ideas and innovations that flowed from it made America a world leader. The nation’s scientific leadership has been essential to our shared prosperity and national security, a

In 2026, MIT granted tenure to 10 faculty members across the School of Engineering. This year’s tenured engineers hold appointments in the departments of Aeronautics and Astronautics, Civil and Environmental Engineering, Electrical Engineering and Computer Science (EECS) — which

Google introduces Gemini 3.5 Flash Cyber, a lightweight cybersecurity model to find and patch vulnerabilities.

🔗 Cũng đưa tin: Google DeepMind

Now you can get more intelligent speech-to-text transcription with Gemini 3.5 Transcribe.

GPT-6 Astra tops Retell’s voice agent benchmark, while Codex helps it reach a $1.6M annual revenue run rate per employee.

With OpenAI Daybreak, Check Point speeds up the research behind new security protections, enabling its analysts to turn vulnerability intelligence into customer defenses faster.

Đây là một hệ thống trí tuệ nhân tạo mới được thiết kế để đọc và hiểu các số liệu về sức khỏe thu thập từ các thiết bị đeo tay của bạn.

Các kỹ sư vừa tạo ra một bộ não nhân tạo thông minh chuyên dùng để đọc và hiểu các bảng số liệu phức tạp mà không cần phải dạy nó từ đầu.

For most languages of the world, language model pre-training operates in a data-constrained regime where models must repeat their training data many times, degrading generalization. Two remedies exist: aggressive hyperparameter tuning such as high weight decay, and mixing in data

Large Language Models (LLMs) are considered to be a more efficient and cost-effective alternative to human judgment for Machine Translation (MT) evaluation. With MT evaluation spanning a large number of language pairs, domains and levels of annotation granularity, LLMs must be th

The increasing demand for GPU-accelerated machine learning inference has exposed limitations in centralized GPU infrastructures, including high cost, limited elasticity, and long wait times under resource constraints. At the same time, modern Apple Silicon–based macOS devices are

On 12 October, Applied Materials unveiled a multibillion-dollar research facility in Sunnyvale, Calif., in the heart of Silicon Valley. The company first announced in 2023 plans to invest US $4 billion in its Equipment and Process Innovation and Commercialization (EPIC) Center, b

On 6 August, the U.S. Government Accountability Office (GAO) released a report on the state of the 2022 CHIPS and Science Act, a bipartisan federal law passed under the Biden administration to boost U.S. semiconductor manufacturing and R&D. While manufacturing projects funded by

Discover how the ZEISS Crossbeam 750 FIBSEM sets a new benchmark for precise TEM lamella prep, tomography, and advanced nanofabrication. This delivers better resolution, better SNR, larger usable FOV, and shorter acquisition times. Learn how uninterrupted FIB milling will reduce

Đây là một bài kiểm tra mới giúp robot thông minh tự động thiết kế hoàn chỉnh một con chip máy tính từ ý tưởng ban đầu cho đến sơ đồ chế tạo thực tế.

Bài viết giải thích cách tinh chỉnh nguồn điện cho các siêu máy tính AI để chúng chạy mát hơn và tốn ít điện hơn giống như việc điều chỉnh lượng nước vừa đủ cho một hệ thống tưới tiêu.

Bài phân tích chỉ ra rằng khi các công ty chế tạo chip chung với nhau, việc giấu kín bí quyết kỹ thuật có thể làm sai lệch thông số dự đoán của linh kiện.

Long-term customer contracts and rising chip prices point to a change in how memory manufacturers manage multi-year production capacity and capex. The post SK Hynix Reports Record Q3 Results Driven by AI Memory Surge appeared first on EE Times.

NSING Technologies, a Singapore-founded semiconductor company, presents this whitepaper on its full-stack semiconductor solutions for embodied AI and robotics, built around high-performance MCUs, security ICs, wireless connectivity and BMS ICs. The document explains the industry

This whitepaper explores how building safety and security systems can overcome the limitations of legacy RS-485 networks by adopting Single Pair Ethernet (SPE). It examines the growing demands of AI-enabled surveillance, access control, cybersecurity, and system convergence, whil

High-level synthesis (HLS) can schedule only the concurrency made legal by its input description. For feedback-coupled designs, a sequential behavioral model can obscure the architectural state boundaries and hazard resolution needed for efficient overlap. We develop an explicit

Hardware components communicate through clock-cycle-sensitive protocols. Hardware Description Languages (HDLs), however, provide limited support for expressing these protocols or for checking that independently developed components agree on them. When components disagree, the com

Analog device sizing must satisfy all specifications across process, voltage, and temperature (PVT) corners to ensure tape-out success, but simulating every candidate at all corners dominates the sizing time and slows design iteration. Conventional surrogate-based optimizers sear

Multi-trigger backdoor attacks, in which heterogeneous trigger mechanisms (e.g., patch, blend, warping, and the latent-domain Spy-Watermark) coexist within a single poisoned model, cannot be addressed by existing shortcut-based defenses that assume a single, static attack mechani

Chiếc xe tự hành trên sao Hỏa đã tự lái đi một quãng đường kỷ lục nhờ vào các phần mềm thông minh tự quyết định đường đi mà không cần con người nhắc từng bước một.

TrackJoint is a Modelica MultiBody component that turns a three-dimensional road surface into a joint with two degrees of freedom. The two joint coordinates are the distance travelled along a reference line and the sideways offset from that line. Together they give the position,

Legged robots for lunar exploration must be designed for low gravity, where reduced contact forces and long flight phases make leg morphology and controller choice interdependent. We study this for a planar monoped in lunar gravity ($1.625$ m/s$^2$) using a bi-level co-design fra

Humanoid dexterous hands are a core component of embodied intelligence, enabling robots to translate high-level intent into rich and adaptive physical interaction. However, existing designs remain constrained by a persistent trade-off among anthropomorphic kinematic complexity, m

Differences in musculoskeletal morphology challenge policy reuse across individual models. Here we present MyoAcross, a framework that combines individual musculoskeletal model construction, morphology-conditioned policy learning and virtual model augmentation to enable muscle-dr

Team NimbRo’s robot in action during the RoboCup@Home competition at RoboCup 2026. Raphael leads the team. Raphael Memmesheimer has been successfully competing in the RoboCup@Home league winning the RoboCup@Home competition five times. He and his colleagues won the best paper awa

Bài viết mổ xẻ các công thức toán học giúp giải thích lý do tại sao các mạng lưới nhân tạo phức tạp có thể tự học hỏi và tìm ra đáp án chính xác một cách ổn định.

Getting in a waymo that hasn't rotated its certstore and getting repeatedly sprayed with the "no hookups" spray by a script kiddie in idaho

138 likes

107 likes

930 points

758 points

470 points

375 points

343 points

318 points

297 points

294 points

250 points

234 points

86 upvotes

78 upvotes

72 upvotes

66 upvotes

51 upvotes

43 upvotes

27 upvotes

Follow up on: https://lobste.rs/s/73ozwc/lobsters_nyc_meetup_on_28th_october Under "Alex". Feel free to come a bit later or earlier. I'll write here updating if we move etc. Walker's is by the Canal St. metro stop: https://maps.app.goo.gl/mdEgSkBV2H1NSXAdA and has food and drinks.

27 upvotes

Các kỹ sư máy tính tìm ra cách giúp các lệnh xử lý toán học chạy mượt mà và tự động điều chỉnh phù hợp với mọi loại chip khác nhau mà không cần viết lại mã nguồn.

Bài viết chia sẻ cách thiết kế các bảng dữ liệu cực lớn trong ngôn ngữ C++ để máy tính có thể tra cứu thông tin nhanh như chớp với hàng tỷ lần mỗi giây.

Bài viết giới thiệu một mẹo toán học thông minh giúp máy tính chia số nguyên lớn nhanh hơn rất nhiều khi làm việc với các hằng số.

Các nhà nghiên cứu phát hiện ra ba lỗi bảo mật nguy hiểm liên quan đến cách mã hóa dữ liệu trong phần mềm máy chủ GoAhead.

Nhóm phát triển ngôn ngữ lập trình Python đã quyết định tạm dừng tính năng giúp tăng tốc độ chạy chương trình vì gặp một số khó khăn về kỹ thuật và bảo trì.

Lập trình viên này tạo ra một hệ thống tiền ảo kỳ lạ chạy hoàn toàn bằng các cửa sổ trình duyệt web trò chuyện trực tiếp với nhau mà không cần máy chủ trung tâm.

SaaS user deprovisioning is one of those responsibilities that's easy to overlook when you're...

53 reactions

Description: Python security, screen capture, spring physics, HiDPI and PyInstaller .exe gotchas. 7...

29 reactions

A visual, 10-minute architectural deep dive into MQTT. Learn how the 2-byte binary wire protocol, QoS 0/1/2 handshakes, topic trees, Last Will and Testament, and MQTT v5 Shared Subscriptions actually deliver IoT telemetry under the hood.

24 reactions

If you're studying computer science right now, or you just graduated, you've probably heard some...

23 reactions

Hoi hoi! 👋   I'm nyaomaru, a frontend engineer and OSS developer working mostly with...

20 reactions

When the Ostrogoths besieged Rome in 537 AD, their first military act was to cut the aqueducts. Eleven gravity-fed channels supplied the city with up to a million cubic meters of water per day, more than enough to sustain a population of over a million, fill the public baths, and

In 2009, the City of Antwerp launched a competition to rethink municipal waste infrastructure — a typology typically relegated to generic asphalt lots and containers. Bovenbouw won with a master plan treating the recycling park not as an industrial backlog, but as essential, high

Designed by Kengo Kuma & Associates in collaboration with Cornelius Vöge Atelier for Architecture and Vilhelm Lauritzen Architects, KAI Water Culture House is set to open on December 5, 2026, on Papirøen (Paper Island) in Copenhagen, Denmark. The 5,800 m² public facility combines

The project is part of a larger architectural complex located within an extensive natural site. The landscape of the wilderness serves as the foundation for the architectural concept.

Pittsburgh's Carnegie Museum of Art has commissioned architecture and urban design practice AD—WO to reimagine the Heinz Architectural Center. The practice, led by Emanuel Admassu and Jen Wood and one of the winners of the ArchDaily 2026 Next Practices Awards, was recognized for

Direct contact with a material allows one to appreciate textures, compare finishes, and observe how its appearance changes under different lighting conditions. At an exhibition, these possibilities are complemented by exchanges with manufacturers and professionals, helping to und

Inhabiting a space involves multiple economic, social, cultural, geographic, political, and environmental factors. Beyond these considerations, however, it engages with processes, boundaries, and possibilities that have repeatedly become subjects of architectural inquiry. "We are

Set within Val de Vie Estate in the Paarl Winelands, House Between Gardens is conceived as a contemporary family home shaped by landscape, climate and the rhythms of everyday living. The architecture reinterprets the Cape vernacular H-plan to create a series of interconnected int

Nazuna Hakone Miyanoshita is a newly opened ryokan in Hakone by Nazuna, a brand originally established in Kyoto. We were responsible for the design of its refurbishment. While Nazuna has traditionally reinterpreted Kyoto culture through the contemporary re-editing of wooden machi

The UMG Campus in Antigua Guatemala represents a critical case study in bridging traditional colonial architecture with massive institutional scale. Developed by ATELIER Oficina de Arquitectura, the primary challenge of this project was adapting a small-scale colonial design to a

Located in the Haras Larissa gated community, near São Paulo, the house was conceived as a retreat from the city in direct contact with the rural landscape.

Tòa nhà chọc trời cao nhất bang Texas đã được xây dựng xong, trông giống như một chiếc ủng cao bồi khổng lồ.

Một đài quan sát màu cam rực rỡ hình chiếc nón đã được dựng lên tại bãi rác ở Nhật để mọi người đến xem và hiểu hơn về cách tái chế rác thải.

Kiến trúc sư nổi tiếng người Trung Quốc chia sẻ rằng các tòa nhà kỳ lạ và sáng tạo chính là cầu nối kết nối trí tưởng tượng với đời thực.

Một tòa nhà trường học mới ở Scotland được làm từ gỗ và đá địa phương để giúp học sinh thư giãn và chăm sóc sức khỏe.

Một ngôi nhà nhỏ ở Chile được thiết kế với mái nhà che rộng lớn và một phần hiên nhà vươn ra lơ lửng ngay trên khe núi.

Công trình nhà tắm công cộng mới ở Đan Mạch được thiết kế với những tòa tháp gạch xếp lớp độc đáo bao quanh các hồ bơi ngoài trời.

Hệ thống biển chỉ dẫn bằng gỗ tự nhiên được thiết kế đồng bộ và hòa quyện vào kiến trúc của một thư viện tổng thống mới tại Mỹ.