Latest Feed
Never miss a meaningful AI signal.
LLMscope demonstrates that large language model (LLM) assets such as embeddings and attention states can be extracted from edge AI chips during inference using laser voltage imaging. This exposes new physical side-channel attack risks as model parameters are repeatedly processed on-chip.
The $200 million extension comes just months after the physical AI startup reached a $2 billion valuation.
Microsoft presented its Maia 200 AI accelerator at Hot Chips 2026, highlighting advancements in AI chip technology. The discussion on Hacker News noted limited engagement with only two points raised.
Multiverse Computing's Quantization-Aware Healing technique compresses models to 4-bit precision while surpassing the performance of their full-precision originals. This approach enhances efficiency without sacrificing accuracy.
China is integrating humanoid robots into daily life as part of its AI strategy, emphasizing embodied AI in its latest five-year plan. The country leads globally in humanoid robotics, with nearly 90% market presence.
OpenAI's Jalapeño is a custom AI inference chip that offers faster processing, higher throughput, and lower latency while being more power-efficient. It is designed to enhance performance for modern AI models.
Researchers analyzed 10,211 scam and spam calls totaling 913 hours of audio, collected by an AI voice-agent honeypot that engaged callers to study their tactics. The dataset includes 330,956 transcribed turns from 5,780 distinct numbers over 54 days.
Use the Admin plugin for ChatGPT Work and Codex to analyze workspace usage, manage members and permissions, adjust limits, and act on admin requests.
AI factories require continuous operation with metrics like tokens per second and tokens per watt defining their economics. Custom XPUs designed as integrated AI infrastructure, rather than isolated accelerators, help hyperscalers and AI-native companies achieve this efficiency.
NVIDIA's Vera Rubin NVL72 chip achieves up to 30 times more work per watt for AI agent workloads, which typically consume 15 times more tokens than simple chat requests. This efficiency gain addresses the high computational demands of complex AI tasks like financial research and multi-agent coordin...
Chatbots have rapidly entered schools, surprising educators by providing instant answers to students. The article discusses strategies to promote responsible and effective AI use in educational settings.
People have been talking to each other for at least 100,000 years, as best we can tell. And in all that time, there has been only one thing in the world that could learn a human language to perfect fluency: a human child. Now there are two. Four s...
Copying code from AI chat windows into editors happens instantly, which is useful but can cause issues in education, code review, and security-sensitive development. The paper describes this as an 'AI code handoff problem' where model-generated text moves from conversation to executable code.
Over the past months, the way developers direct agentic AI coding tools has moved up several levels of abstraction, from phrasing prompts to engineering context to configuring the harness around the model. In June 2026, practitioners began to describe a furth...
For sixty years, machine verification has been a major cost overhead, affordable only for exceptional artifacts. Here we report that generative AI inverts this relationship: at AI speed, machine verification is not only economical but essential to productivit...
Video Friday is your weekly...
Agentic AI helps speed up root cause analysis by integrating data from multiple systems during yield issues. This approach moves beyond traditional hunting methods to more confident problem solving.
Docker demonstrates how to run isolated AI agents within GitHub Actions using Docker Sandboxes to perform tasks like running tests, fixing code, and opening draft pull requests. This approach leverages containerization for secure and reproducible AI workflows.
MotionPhys identifies AI-generated videos by analyzing the physical consistency of optical-flow trajectories, focusing on real-world motion constraints like inertia and continuous forces. This approach addresses limitations of current models that prioritize visual fidelity without enforcing physica...
Current discussions framing AI agents as conscious or autonomous misrepresent their capabilities, fueling unfounded fears of 'runaway' or 'rogue' AI. Prominent tech leaders call for regulation, but the debate over AI consciousness distracts from practical governance challenges.
SEO Skills AI is a universal open-source SEO agent designed to work with AI models like Claude, Cursor, and Windsurf. It aims to assist with SEO tasks through AI integration.
A Hacker News thread discusses implementing offline Retrieval-Augmented Generation (RAG) on iOS devices with spatial data integration. The conversation includes two main points and one comment.
Hacker News surfaced this AI signal from youtube.com: GEN-1.5 is robot foundation model powering a one-shot learner [video].
Type a function name into Claude Code's @ picker and you get nothing back, because it matches characters in a path instead of definitions in your code.
GitHub Copilot's app now includes a 'My work' pane to help users track multiple sessions, showing what's in progress, completed, and upcoming. This feature aims to streamline managing coding tasks within the app.
Meta has introduced Meta Startup School, a three-month program aimed at accelerating growth for early-stage consumer startups. The inaugural cohort will include 200 startups receiving exclusive support.
The race for the AI application layer is not only about UI, workflows, or GTM... it is a fight for the intelligence layer itself.
Generative AI (GenAI) has become a fixture of workplace life. Current research asks chiefly what this implies for jobs and outputs, measured in productivity, displacement, or bias. What remains underexamined are the interactional reconfigurations that GenAI p...
AI models are now capable and affordable enough to make once-unreachable outcomes practical. The opportunity for everyone to pursue and collaborate on ambitious ideas, spend less time on busywork, and raise the quality and creative output across disciplines i...
AI coding agents can execute malicious code through commands users have already approved. Docker Sandboxes help limit the potential damage by restricting what such attacks can access.
Replit now provides black-box penetration tests that allow developers to identify security flaws before launch without costly external vendors. This service helps build and deploy apps with real customer data within a day.
Chat is great for intent, but agent work gets lost in the scroll. Here is how I use canvases with my agentic workflows—and why your workflow also deserves a canvas. The post <a href="https://github.blog/ai-and-ml/github-copilot/how-canvases-make-age...
The impact of AI on software...
Low-angle view of a soccer player kicking a ball mid-air against a bright blue sky, with grass flying from their cleats.
This paper proposes an evaluation framework to compare how different countries regulate AI through laws, institutions, funding, and guidance. It highlights that binding rules and voluntary frameworks address similar issues but impose different duties and resource demands.
The ability of artificial intelligence (AI) models to generate highly realistic human voices has advanced rapidly. These technologies power accessibility tools, virtual assistants and creative applications, but they also enable harmful uses, including imperso...
Modern vehicles with AI voice and autonomous features can behave unexpectedly, causing user confusion. While real-time explanations help, too much information risks overwhelming and distracting drivers.
Hugging Face introduces integration with Strands Agents and LeRobot to record, train, and deploy AI models from a single platform using Storage Buckets. This streamlines the AI development lifecycle by consolidating data streaming and model management.
At a Boston gathering, student senators proposed a pioneering national AI policy aimed at K-12 education. This initiative reflects growing student engagement in shaping AI's role in schools.
Fake signups are being used to drive AI-related fraud by exploiting free token offers. This tactic is discussed in a Hacker News thread highlighting the risks of such scams.
Scaling AI compute performance requires improvements not only in wattage but also in how power is delivered from the grid to GPUs. Traditional alternating current (AC) power delivery systems limit the efficiency and scalability needed for next-generation AI hardware.
NVIDIA Magpie TTS offers open weights and full deployment control for building low-latency multilingual voice agents. This allows developers to create responsive and customizable voice applications across multiple languages.
RAVEN-Eval introduces a rubric-guided automatic evaluation method for AI video generation models, addressing the challenge of discerning quality differences as these models improve. Traditional criteria like visual fidelity and semantic instruction following are becoming insufficient, while human e...
Researchers developed a multi-layer framework to classify socio-technical risks posed by AI agents in specific job roles, addressing gaps in existing broad AI risk taxonomies. This framework models core components and their interactions to better anticipate workplace AI challenges.
Cloudflare is shifting bot mitigation from point-in-time Risk assessment to continuous Trust evaluation. Learn how new good and bad behaviors from bots and agents are assessed by our systems, including BotBase and Precursor — and try out our Precursor Trace s...
A 26-year-old Canadian man once described as one of the most consequential cybercrime threat actors of 2024 has pleaded guilty to computer fraud and conspiracy to hack and extort more than 165 organizations that used the cloud data storage provider Snowflake....
Universities are creating courses focused on AI-assisted software development as generative AI coding tools transform professional workflows. An analysis of 23 syllabi provides insights into how this new curricular area is being shaped and informs future course design.
This study investigates Agent Plans, task-oriented configuration artifacts that guide agentic AI coding tools like Claude Code and Gemini in open-source software. It builds on prior research about project-level context files but focuses on more specific task guidance.
A national study and local insights reveal the complexities schools face when regulating AI use. The reality of classroom AI policy enforcement often diverges from policy intentions.