Hottest toy for Christmas? Nvidia CEO plays with adorable Microduck robots, 20,000 sell

The robots, stomping their little feet and making tiny R2D2 noises, sat, fell down dead for a "nap," and summersaulted on voice command.

Kimi K3.1 Pro and K3.1 Pro Highspeed

Gemma-4-26B-A4B with 39-69% less thinking at same accuracy

Vincent van Gogh's masterpiece "The Starry Night" may depict a real night sky.

Vincent van Gogh's masterpiece *The Starry Night*, painted in 1889, may depict the actual night sky over France on July 22, 1889. *The Starry Night* is widely believed to be a stylized representation of the view from Van Gogh's window in Saint-Rémy-de-Provence, France. The controversy lies in whether this night sky was the post-impressionist painter's own fantasy—a symbolic expression—or a result of genuine astronomical observation. Furthermore, the exact date of the painting's creation remains undetermined. To address this issue, researchers at Kyung Hee University in South Korea have developed a...

FlagOS 2.2 by Zhiyuan Zhongzhi enables the synergistic evolution of cutting-edge models and diverse computing power.

On October 17, 2026, the 2026 Artificial Intelligence Open Computing Conference and Zhongzhi FlagOS Technology Conference, themed "One Stack for Seamless Connectivity, Intelligent Computing Without Borders," was held in Beijing. Following the release of FlagOS 2.2 on September 30, the conference showcased technical advancements spanning from low-level operators and compilers to model deployment and services, while highlighting the pressing reality that AI models undergo rapid iteration cycles whereas computing infrastructure requires long-term service. Zhongzhi FlagOS 2.2 has evolved from multi-chip adaptation to cross-chip reuse, with its nine operator libraries collectively covering 1,454 operators, an increase of 890 from the previous version...

Featuring guests such as Liu Tiyan, Nobel laureates, and Turing Award winners, they discuss AI's next decade: What is AI's next ace card after the dominance of Transformers?

In 1956, the Dartmouth Conference coined the term "artificial intelligence" for this field of study. Seventy years later, as AI has moved from laboratories into countless industries, a new question faces researchers: If machines can truly learn, reason, write code, and even contribute to scientific discovery, where will artificial intelligence head in the next decade? At the first roundtable discussion of ICAAS 2026, participants include Michael Levitt, Nobel Laureate in Chemistry and Professor at Stanford University; Jack Dongarra, 2021 Turing Award winner; mathematician Ma Xiao; and Zhongguancun Academy...

PostgreSQL 19 failing to release? AI "spot-checking" is too fast; humans can't keep up with fixing bugs.

Recently, the release of PostgreSQL 19 encountered an unusual setback. The PostgreSQL team announced that the release, originally scheduled for October 29, will be delayed to November 19. Prior to this, an additional RC2 candidate version will be released on November 12. The official reason given is that a significant number of fixes have recently been added to the maintenance branches, and the team hopes to first have the community validate these fixes through the RC2 release before proceeding with the official release. Meanwhile, after the Beta phase began, approximately 8 to 10 important features were...

The UN warns that El Niño could lead to food shortages for tens of millions of people.

This year's El Niño event could reach unprecedented intensity. It is projected to peak around late 2026, but its impacts will persist for a long time thereafter. Sea surface temperatures in the tropical Pacific (the core region of El Niño) may exceed the average by 3.8°C between November 2026 and January 2027. A very strong El Niño can disrupt normal temperature and precipitation patterns across many regions. It could bring severe droughts to some areas, heavy rainfall and flooding to others, and cause temperatures to rise across vast parts of the world...

Global academic journal evaluation is undergoing a new transformation: shifting from single indicators to diverse criteria, with China's evaluation system now joining the international landscape.

For a long time, researchers have inevitably relied on metrics such as the impact factor, SCI inclusion status, and journal quartiles when selecting journals for publication, while academic institutions have depended on these same indicators to evaluate scholarly outputs. Often, where a paper is published receives more attention than what the paper itself studies or the problems it solves. However, as global research activities become increasingly diversified, this evaluation system, which relies heavily on a limited number of international databases and assessment metrics, is facing new challenges. Different disciplines have distinct publication habits, and regional journals and emerging research fields...

Why is Chery considered an AI company?

Why is Chery an AI company? It has rebuilt its car-making process using AI, transformed this capability into a reusable foundation, and enabled the emergence of new species beyond automobiles. Author | Wang Ruihao Editor | Li Yuchen "Several executives told me that Chery's cars are increasingly less like cars, and Chery itself is becoming less like an automotive company. This is exactly what I wanted to hear—Chery is a high-tech innovation company merely wearing the disguise of an automotive enterprise." On October 20, the 2026 Chery Global Innovation Conference opened in Wuhu, Anhui. The words behind this statement...

GitLab 19.4 Released: New Goal-Driven Agents and Enhanced Cost Control

GitLab releases DevSecOps platform version 19.4, introducing a goal-driven mode for GitLab Duo CLI, three hosted open-weight models, additional Model Context Protocol (MCP) tools, and more granular AI cost control capabilities. This release expands GitLab's agents beyond single prompts and predefined workflows, while agent actions remain constrained by the project's existing permissions and approval rules. The newly added `/goal` command in GitLab Duo CLI is currently in public preview; it accepts an open-ended goal and continues to execute...

Decision models now have dedicated inference engines: StartLux open-sourced Prism, achieving up to 5.3x speedup on a single GPU.

Recently, two Chinese AI companies have successively open-sourced inference systems optimized for their own models. On October 20, DeepSeek open-sourced the inference framework DeepInfer tailored for its own model. Subsequently, StartLux also announced the open-sourcing of StartLux-Prism, a dedicated inference engine designed for StartLux-Decision. Although the two companies follow different technical routes, both have extended their optimization focus from the models themselves to the inference systems. According to the test data released by StartLux, using the StartLux-Decision v1.1 27B model on a single NVIDIA H200 GPU, Prism...

Q&A: How AI came to model the “messy” human mind

Assistant Professor Benjamin Lindquist studies how engineers deliberately injected randomness into early AI, giving rise to deep-learning models that sometimes hallucinate.

"PHP2Shell" Rings the Alarm Bell for Enterprise Vulnerabilities, QiAnXin TianJi Offers a One-Click Rapid Check Solution

On October 21, the globally widely used server-side programming language PHP was found to have two critical security vulnerabilities (CVE-2026-109618, CVE-2026-109619), collectively known as "PHP2Shell." Verified by Qianxinan technology experts, these vulnerabilities enable unauthenticated remote arbitrary command execution, representing a severe internet infrastructure-level security risk with potential impact on over a million assets in China. Due to the extremely low exploitation threshold, they are highly likely to evolve into large-scale worm-like propagation attacks, directly threatening the security of China's critical information infrastructure, government and enterprise websites, and...

DeepSeek's multi-billion-dollar financing is nearing completion: by allying with industrial investors such as Perfect World and SMIC Juyuan, the company is doubling its fundraising to bet heavily o...

Just under four months after DeepSeek's first round of financing, which set an industry record at 51 billion yuan, closed, the second super-round financing for this leading large-model enterprise is nearing completion. According to the latest report from Bloomberg today (October 22), the scale of DeepSeek's current financing round has been locked in at approximately 100 billion yuan, with a pre-money valuation remaining around 500 billion yuan. Insiders reveal that the deal underwent significant dynamic adjustments during its progression: the initial planned fundraising amount was 50 billion yuan, followed by...

Award-winning AI literature

Show HN: agent-usage-guard – a Claude Code mod to protect your usage window

Causal Chain of Thought: Mitigating Attention Leakage via Context Isolation

Chain-of-thought (CoT) prompting elicits step-by-step rationales that appear to explain model predictions, but there is no guarantee these steps causally drive the final answer. In standard autoregressive generation, dense attention allows each reasoning step to attend to the full context, enabling the model to derive answers through direct lookup while generating plausible-sounding but post-hoc justifications. We introduce CausalCoT, an inference framework that enforces causal transparency b...

Mini brain organoids ‘reading’ Braille key step towards biohybrid computing

Training Small Language Models for Transferable Tool Selection

As LLM agents operate over corpora of more than ten thousand tools, selecting the right tool for a complex multi-step user instruction becomes a central bottleneck. Tool descriptions cannot all fit in context, and relevance-based retrievers only rank tools without knowing which are actually needed. Recent work instead leverages proprietary commercial LLMs like Claude or ChatGPT to drive a retriever, decomposing the instruction and selecting over retrieved candidates. We show that a small, ope...

Running Frontier-Class AI Models Across 8 NVIDIA DGX Spark Nodes

AI developers aren’t just running a single model anymore—they’re orchestrating multiple models in parallel, chaining inference calls, and experimenting...

Show HN: Run full iOS 16.5 firmware on Android Devices

What’s getting pwned right now

Study Finds AI-generated Code Fails to Boost Productivity Due to Downstream Bottlenecks

A recent Harvard study of 718 firms, using the Jellyfish analytics platform, finds that although AI coding assistants and agents increase code volume, this does not translate into a corresponding increase in completed work, partly because AI-generated code increases the workload in later stages of development, particularly code review. By Sergio De Simone

Can AI Co-Design Structures? An AI Agent for Early-Stage Structural Design

Recent advances in large language models and agentic workflows offer a new paradigm to early-stage structural design translating design intent into structural models. Although text-to-3D approaches have demonstrated that natural language can be used to produce visual, mesh-based geometric models, structural design requires domain-specific geometric arrangements and semantically rich data types. This paper introduces a proof-of-concept workflow for prompt-to-structure generation using an AI ag...

Another domestic decision model is released: designed for high-frequency judgment tasks in enterprises, its 23 open-source benchmark accuracies surpass those of Microsoft and JEV.

On October 22, Ant Group's Digital Technology Company released the decision model LingDT-Decision-1.0, offering two parameter scales: 4B and 27B. Primarily targeting enterprise application scenarios such as intelligent customer service, security risk control, and finance, this model differs from general-purpose large language models that focus on text generation. Instead, it mainly handles structured decision tasks such as intent recognition, task classification, tool selection, and risk assessment. The model is now available on the Ant Group Digital Technology Company's DTMaaS platform, where developers can enjoy a limited-time free trial. According to test results published by Ant Group's Digital Technology Company, in certain enterprise business scenarios, LingDT-...

In-depth Analysis of the Global 72-Hour Security Storm: The PHP Incident Was Only the First Shot

If you happened to fall into all of these traps yesterday, you're probably the most unfortunate security engineer in 2026. Imagine your website running on Linux servers, with Nginx handling reverse proxying for the frontend, a mixed backend using Go, PHP, and Node.js, code stored in your self-hosted GitLab, updates pushed via an automated release pipeline, and HTTPS certificates issued by eMudhra. Congratulations, you've just acquired the full suite of security vulnerabilities. On October 21, the PHP community disclosed two remote code execution vulnerabilities with a severity score of 9.2 out of 10 (CVE-2025-41661 and CVE-2025-41662)...

As enterprise AI becomes increasingly complex, how can a single Gateway effectively manage models, agents, and costs? | Technical Trends

In 2026, what substantive breakthroughs will agents achieve in enterprise applications? Download the "2026 AI and Data Development Forecast" white paper to gain expert insights and embrace new ways of working ahead of the curve! Over the past few years, the enterprise AI technology stack has evolved into a highly complex yet fragmented and difficult-to-manage ecosystem. Today, this stack encompasses components such as vector databases, monitoring systems, and pipeline optimization, along with programmable agents, embedded agents, and models integrated with enterprise systems...

SecLog-Bench: A Scalable Benchmark for Cybersecurity Log Reasoning with Scaffolded Evidence

Large language models (LLMs) have shown strong potential for cybersecurity, leading to growing interest in LLM-based automated cybersecurity log analysis. However, current datasets (1) are limited in scale; (2) rely mainly on single-source evidence; and (3) have only binary or multi-class attack labels, without dense annotations of the reasoning process. To address these limitations, SecLog-Bench has three key features. First, we build an extensible simulation platform that covers 29 attack s...

High-Performance Optimization Practices for Linear Attention in the Qwen Series of Models

As large model sequence lengths advance from tens of thousands to millions, the O(N²) memory and computational overhead of softmax attention has become a stringent constraint. In recent years, mainstream models such as Qwen and Kimi have gradually shifted toward a hybrid architecture combining "linear attention + softmax attention." However, efficient GPU implementations of linear attention still face systemic challenges including insufficient parallelism, redundant memory access, and excessive SFU (Shared Function Unit) load. This paper is compiled from a presentation by Zhang Ruidong, an algorithm engineer from Alibaba's Qwen team, at AICon 2026 in Shenzhen, titled "Linear Attention for the Qwen Series Models..."

Enterprises have the demand, the FDE team has the capability—so why is it still difficult to implement AI projects?

On October 23, 2026, at the opening ceremony of QCon Shanghai 2026, Huo Taiwen, founder and CEO of KeGang Technology, officially announced that the company's mission has been upgraded to "Empowering enterprises to accelerate the implementation of AI." At the same time, ModelArts has comprehensively upgraded from an AI community platform to an enterprise AI implementation service platform, creating an "AI Supply-Demand Brain" that connects real enterprise needs, professional AI delivery teams, and high-quality AI products. "The bottleneck in AI implementation has never been technology; it is trust and information," Huo Taiwen pointed out in his speech, noting that when enterprises seek reliable products...

Tencent's Yao Shunyong's AI "Overtime Theory"

At the AI poker table, emotions are becoming extremely anxious.

Developer Behind the AI-Built Adobe Clones Says They Are “Not Anywhere Close to Ready

A developer used Claude Opus 5.5 to build open-source replacements for seven Adobe applications, promoted them across Reddit with a projection of roughly 100% parity within a month, then told Hacker News the project was not anywhere close to ready. Its own roadmap puts it at 45% ready for real work and 2,300 to 4,100 agent-hours from full parity. By Steef-Jan Wiggers

AI has turned the entire software ecosystem into a sieve; it is the darkest hour for full-stack developers.

Authors: Lin Qibei, Tina Editors: Cai Fangfang If we were to summarize this magical week, it would be something like: "The darkest hour for full-stack developers, a cyber survival of the fittest in the AI era." Imagine you are an ordinary developer; as soon as you arrive at work in the morning, you discover that the entire internet infrastructure is failing one after another. GitLab exposed a critical vulnerability with a severity score of 10.0, Chrome urgently patched a high-severity vulnerability, Node.js and PHP were hit back-to-back with major issues, LiteLLM suffered a supply chain attack, eMudhra's certificate private key was leaked, and even the cloud deployment platform Railwa...

TESSA: A Governance-Oriented Custom GPT for SME Guidance under the EU AI Act: Design, Implementation, and Evaluation

The increasing adoption of generative AI presents new opportunities while raising AI governance and regulatory challenges, particularly for small and medium-sized enterprises (SMEs) navigating the European Union Artificial Intelligence Act (EU AI Act). To support AI governance education, this study developed TESSA, a governance-oriented educational artifact within the OpenAI Custom GPT environment. Following a design science research approach, TESSA was evaluated against baseline ChatGPT usin...

GitHub migrates Copilot runtime to Rust with AI-assisted refactoring

GitHub has migrated the runtime behind GitHub Copilot CLI, Copilot apps, and the Copilot SDK from TypeScript and Node.js to Rust, replacing over 800,000 lines of code through AI-assisted refactoring. This migration took approximately 14.5 weeks and involved 128 pull requests; during the migration period, GitHub continued to release the runtime version. GitHub reports that for client startup, session creation, and single-turn interaction scenarios, the legacy runtime took 5.25 seconds, while the Rust runtime, with in-process embedding, reduced this to 292 milliseconds. This migration also changed the application...

Volcengine launches the first batch of enterprise trials for the Seedance 2.5 4K model, exclusively supplying 20 global studios and directors.

Recently, Volcengine launched the first batch of enterprise user trials for the Seedance 2.5 4K model, binding it to specific works and providing targeted access to 20 global studios and directors to explore high-end cinematic production needs. Beyond enhanced visual clarity, the Seedance 2.5 4K model emphasizes more nuanced and natural character performances, along with finer consistency in reference-based generation and editing, making it the strongest model in the Seedance 2.5 generation. A film studio that has already tested the model told Leifeng.com that the pricing structure of Seedance 2.5 4K is not as expected...

Google staff resign in backlash over its AI contracts with military

Google's Rust Style Guide

Canary Tokens

Mitigating Model Collapse through Dreaming Learning

Training generative models on synthetic data causes loss of diversity and a decrease in performance, a phenomenon known as model collapse. We study this degeneration spectrally in a controlled Markov setting. The eigenvalue spectrum of the learned transition operator offers a direct view of the process, since the subdominant eigenvalues set how fast the chain mixes and therefore how much of the state space it continues to explore. We find that sampling at temperature $T_s 1$. Because high-tem...

Today's PHP security release, and how we're strengthening security

Birds are one of the most important nutrient transporters on Earth.

The role of birds in shaping the geochemical environment has been underestimated; they are among the most important nutrient transporters on Earth. Researchers have revealed how birds move nutrients between land, lakes, and oceans through their excrement and bodies. This process profoundly influences everything from the chemical properties of soil and water bodies to the growth of plants and animals. Their ability to fly allows birds to traverse vast distances, even crossing natural barriers such as oceans and mountain ranges. Birds are also highly efficient in processing and cycling nutrients. The nutrient intake rate of birds compared to their body size...

Conversation with Amazon CTO: Still Hiring Junior Engineers, Focusing on China's Open Models

On October 14, I participated in a media group interview with Werner Vogels, CTO of Amazon, and asked him a question about people: With AI becoming increasingly proficient at writing code, what about fresh graduates and junior engineers? In my view, Werner is not an ardent proponent of AI—he has consistently been cautious and has always emphasized the value of humans. However, whether companies will continue to hire new talent is a more objective choice. A truly rational approach would reflect its judgment on AI in actual hiring actions. "The number of junior engineers we recruit is...

Capital Online: Turns Profit on Books, Continues Investment in AI Computing Power | Financial Report Review

Maori restoration underway, but funds remain tight.

Presentation: AI is an Amplifier: Scale High Performance, Not Dysfunction

Nathen Harvey shares insights from DORA research on AI adoption in software delivery. He explains why AI acts as an amplifier, how the DORA AI Capabilities Model drives outcomes, and why engineering leaders must evolve beyond bottlenecks like code reviews to achieve sustainable productivity, reduce delivery instability, and empower developers without burning out their teams. By Nathen Harvey

Open Inference Code Narrows the Gap Between Rival AI Chips

DeepSeek released serving software that builds for Huawei accelerators and Nvidia GPUs from one codebase. In China, where boards are scarce and token traffic is rising, that choice could reshape how teams pick hardware—if live traffic backs up the demos.

Are AI Agents Accelerating the Smartphone Boom? The Dilemma of Device Replacement, Edge AI, and the Battle for the Next-Gen Entry Point

The mobile phone needs a "new story."

Is Object Storage "Eating" Kafka? Cloudflare Rebuilds Event Streams with R2

Cloudflare recently launched the K2 public beta. This is a serverless event streaming service built on top of partitioned persistent logs running in R2 object storage. Trading latency for cost advantages and elasticity, Cloudflare also released specific metrics: the p99 latency in production is approximately one second. However, public testing reveals that its end-to-end delivery speed is slower than this figure. The service was initially developed as a foundational pipeline for "Basin Pipelines," which requires a persistent location to stage events while waiting for their pull-based processing...

Signals and Queries: Token-Efficient Agentic Context Discovery

User-facing LLM agents must identify which behavioral policies apply at each turn, but evaluating every policy with an LLM is costly. We introduce Signals and Queries (S&Q), a lightweight first-stage relevance filter developed for Parlant 4.0 that removes at least 30% of candidate policies before downstream evaluation while preserving high recall. For each policy, an LLM generates activating examples and hard negatives offline. An independent logistic classifier trained on their embeddings th...

Blog Review: Oct. 21

From server to rack; AMBA CXS; digital twins; edge AI development loop; satellite power; global talent challenges. The post Blog Review: Oct. 21 appeared first on Semiconductor Engineering.

Cloudflare has open-sourced Clef, a decision model for AI agents.

During the recent "Birthday Week" event, Cloudflare released Clef, a set of open-weight AI models designed to make choices among predefined options rather than generate text. Cloudflare released versions of the model with 9 billion and 27 billion parameters, along with a platform to adapt them for specific decision-making tasks. Clef is a 27-billion-parameter multimodal model that takes states and a Schema with typed questions as input and outputs decision results. It can process text, JSON, images, or videos with a single forward pass...

DeepSeek releases its latest open-source inference framework: single-cluster daily throughput of 155 trillion tokens, with KVCache capable of retaining data for 7 days.

On October 20, DeepSeek announced the official open-source release of the large-scale cloud service inference framework, DeepInfer. This framework is already deployed in large-scale online inference services for DeepSeek V4.1 Flash. According to DeepSeek, DeepInfer integrates high-performance operators and a deeply optimized inference engine, supporting capabilities such as distributed inference, speculative decoding, SWA Bounded Replay, and structured outputs. It is capable of running on both NVIDIA and Huawei Ascend platforms. Based on the data released by DeepSeek, the throughput of a single cluster service running on this system has already reached 155 trillion tokens per day.

Daily Brief: US Government directs agencies to discontinue use of DeepSeek; Li Auto unveils its self-developed "Lianshan" three-electric powertrain system.

The Ministry of Industry and Information Technology solicits opinions on five mandatory national standards, including automatic parking; Mastercard and Apple Pay add three partner banks, among other updates. View full article

Anthropic’s cloud partnerships are getting messy

Anthropic is reportedly struggling to secure enough compute from cloud providers that increasingly view it as a competitor, even as its infrastructure spending balloons ahead of a planned November IPO.

Automating ML for Science: Can Frontier Agents Climb Scientific Hills in the Wild?

Frontier coding agents are increasingly framed as end-to-end automated researchers, and ML for science offers a demanding test of that promise: improving a scientific model often requires domain insight beyond routine ML engineering. We introduce Agent4Sci, a long-horizon testbed for measuring how far today's agents can improve the published best models on scientific ML benchmarks, spanning 20 benchmarks across 11 domains. Rather than ranking agents on a leaderboard, Agent4Sci asks what kinds...

ByteDance, Alibaba, and Tencent have ordered over $35 billion worth of RTX PRO 5500 graphics cards.

ByteDance, Alibaba, and Tencent have placed orders for over 3 million NVIDIA RTX PRO 5500 workstation graphics cards, valued at more than $35 billion. NVIDIA has informed its customers and partners that, due to demand from Chinese tech companies far exceeding planned production, it cannot guarantee fulfillment of all orders. NVIDIA told partners that the actual quantity they may receive could be as low as 5% to 10% of what was ordered. NVIDIA's global production cap for RTX PRO 5500 graphics cards in 2027 is 3 million units. Orders from Chinese customers...

AI-generated slop games are flooding Steam's Summer Sale.

The Steam Next Fest kicks off on October 19, and players immediately noticed it's flooded with a vast number of AI slop games. What was once a game event renowned for uncovering hidden gems has now devolved into a showcase for cheap AI-generated games. Players are calling on Valve, the operator of the Steam platform, to introduce an AI filter feature that would allow users to filter out these AI-generated games with a single click. With the advancement of AI tools, reverse engineering has become easier than ever, leading some game developers to be reluctant to release free demo trials, fearing that DE...

AI agents are about to put least privilege to the test

Presented by JumpCloudRead more

Google and Synaptics unveil new AI chips for wearables and industrial devices, powered by open-source Coral technology

Google is taking another step toward moving AI beyond the data center and into the devices people use every day, partnering with semiconductor manufacturer Synaptics on a new generation of processors designed to run AI models directly inside smart glasses, cameras, industrial sensors and other connected devices.Read more