Search the archive

live to code along

VSCAN

VSCan is a tool that analyzes Visual Studio Code extensions for security vulnerabilities. You can input an extension's name or ID, and VSCan performs a deep analysis of its code, permissions, and metadata. The tool then generates a report detailing potential risks and recommendations.

AUTUMN

Autumn is an open-source platform that makes pricing and billing easier by acting as an intermediary between Stripe and your application. It supports various pricing models like subscriptions, credit systems, and usage-based models, without requiring extensive webhook management.

INTL-T

Intl-T is a fully-typed, object-based i18n translation library for TypeScript with features like autocomplete, node-based translations, and support for JSON files and dynamic remote imports. It integrates easily with React and Next.js and provides components for React injection and server-side rendering.

THE BITTER LESSON IS COMING FOR TOKENIZATION

Tokenization-free LLMs are possible. Tokenization is efficient but can lead to information loss and various issues that have to be addressed later on with different methods. The post explains recent advancements in byte-level models like ByT5, MambaByte, MEGABYTE, SpaceByte, and ends with a detailed analysis of the BLT architecture, which uses an entropy-based dynamic patcher to compress sequences and allocate compute more efficiently.

BASIC FACTS ABOUT GPUS

This article goes over the intricacies of GPU performance, focusing on maximizing efficiency through understanding the compute and memory hierarchy. It explains how to analyze kernel performance, categorized as memory-bound, compute-bound, or overhead-bound, and introduces the Roofline model for visualization.

MCP IS EATING THE WORLD—AND IT'S HERE TO STAY

MCP is gaining traction despite previous attempts at connecting LLMs to external tools falling short. This is because MCP benefits from better LLMs, a vendor-neutral protocol design, and straightforward tooling, making it easier for developers to adopt it for building and sharing tools.

(MAP LLMS EXCEL) SAVED MY PASSION FOR GAME DEV

Frustrated by tedious data entry in his game development project, this developer rediscovered his passion by using LLMs to map his Excel data into C# code, saving him time and preventing burnout.

GEMINI ROBOTICS ON-DEVICE BRINGS AI TO LOCAL ROBOTIC DEVICES

Gemini Robotics On-Device is a new, efficient, and adaptable on-device robotics model optimized for general-purpose dexterity, fast task adaptation, and operation in environments with limited connectivity.

INFRASTRUCTURE AS CODE IS A MUST HAVE

Infrastructure as Code (IaC) means defining infrastructure configurations as code (e.g., using Terraform) - it is necessary for modern cloud environments because it allows for reusability, easy rollbacks, transparency, and faster deployment.

CONNECT ANY REACT APPLICATION TO AN MCP SERVER IN THREE LINES OF CODE

Cloudflare's use-mcp is a React library that makes the connection of React applications to MCP servers easier with just three lines of code.

ZUCKERBERG LEADS AI RECRUITMENT BLITZ ARMED WITH $100 MILLION PAY PACKAGES

Mark Zuckerberg has personally reached out to hundreds of researchers, scientists, infrastructure engineers, product stars, and entrepreneurs to try to get them to join a new Superintelligence lab. He is offering them hundreds of millions of dollars and in at least one case discussed buying a startup outright. Despite the financial incentives, some potential candidates have been hesitant to join Meta. It is unclear how successful Zuckerberg's efforts will ultimately be - OpenAI has given counteroffers to people Meta has tried to poach.

COURT FILINGS REVEAL OPENAI AND IO'S EARLY WORK ON AN AI DEVICE

Legal filings submitted as part of a trademark dispute reveal details about OpenAI and io's efforts to build a mass-market AI hardware device. The companies have been vigorously researching in-ear hardware devices. However, io's chief hardware officer said in court that the prototype OpenAI CEO Sam Altman missioned in io's launch video was not an in-ear device nor a wearable device. The design of the prototype has not yet been finalized and the product is at least a year away from being advertised or offered for sale.

NEW YORK TO BUILD ONE OF FIRST U.S. NUCLEAR-POWER PLANTS IN GENERATION

New York plans to build a large nuclear power facility to add at least 1 gigawatt of energy, enough to power about a million homes. The New York Power Authority will find a site in upstate New York and determine the reactor's design. It may pursue the project alone or in partnership with private entities. The US nuclear industry has struggled for more than three decades to do much more than manage existing plants because of safety concerns, huge cost overruns, and a permitting process that moves at a glacial pace. This project could help jump-start a new era of US nuclear reactor construction.

A PRIMER ON THE CURRENT STATE OF LONGEVITY RESEARCH

This post looks at the state of affairs in aging research. It offers some basic background knowledge for each topic covered. Longevity-focused research hasn't led to anything significant yet, but the field is still very young. Scientific discovery works in strange ways - research from longevity could end up impacting other fields, leading to unexpected returns.

BigPicture Enterprise. Take the interactive tour

PICKAXE

Pickaxe is a TypeScript library for building fault-tolerant and scalable AI agents. It handles the complexities of durable execution, queuing, and scheduling so that developers can focus on writing core business logic. Pickaxe enables developers to build agents that call tools, other agents, or any other functions they define. A video demo is available.

PYTHON CAN RUN MOJO NOW

Python can call Mojo code now. The setup is pretty simple - this post walks through how to set it up and provides some examples of how to use it. Development on the stack is still early, so it's not yet production-ready. Mojo is a lot easier to learn than Rust while still providing a meaningful function speedup.

CHECKING IN ON AI AND THE BIG FIVE

This post evaluates the state of AI's biggest players, primarily through the lens of the Big Five: Apple, Google, Meta, Microsoft, and Amazon. It looks at each company's infrastructure, models, partners, data, distribution, core business, scarcity risk, and more and analyzes its position in the industry. The post also looks at model makers like OpenAI and Anthropic and discusses the elephant in the room: China.

THE AI LIFESTYLE SUBSIDY IS GOING TO END

Venture capital and zero-interest-rate policy supplied a ton of money to startups, allowing them to grow even if they were selling customers services at a loss. Today, investor money is going straight into subsidizing the costs of AI products. AI products will likely soon start showing ads to monetize. AI serves its users for now, but in the near future, AI will serve ads.

APPLE HEARD YOUR COMPLAINTS ABOUT THE LIQUID GLASS CONTROL CENTER

Apple has updated the Liquid Glass Control Center to be a bit more opaque in the second iOS 26 developer beta - a comparison is available in the article.

LOGTAPE

LogTape is a logging library for JavaScript and TypeScript that provides a simple and flexible logging system.

SAM ALTMAN OPEN TO ADS ON CHATGPT, CALLS INSTAGRAM ADS ‘KINDA COOL'

Sam Altman previously described advertising as a last resort for OpenAI, but his stance has since softened.

DISNEY AND NBCUNIVERSAL'S MIDJOURNEY LAWSUIT ISN'T ABOUT MONEY — IT'S ABOUT SETTING AI PRECEDENT

The Midjourney suit has the potential to set a precedent around artificial intelligence and how AI companies can operate or train their models - it will have an impact on all creatives.

GOOGLE BRINGS NEW GEMINI FEATURES TO CHROMEBOOKS, DEBUTS FIRST ON-DEVICE AI

Newer, more powerful Chromebooks will soon have image generation, text summarization, and more built into the OS.

APPLE TO ANNOUNCE MORE APP STORE CONCESSIONS FOR THE EU

Apple is preparing to offer new concessions primarily related to its steering rules ahead of a June 26 deadline, after which the European Commission can impose escalating daily fines of up to 5% of Apple's average global turnover.

Explore the latest tools and technical guides

COURT FILINGS REVEAL OPENAI AND IO'S EARLY WORK ON AN AI DEVICE

Court documents submitted as part of a trademark dispute revealed new details about OpenAI and io's efforts to build a mass-market AI hardware device. The companies have been researching in-ear hardware devices, but the prototype OpenAI CEO Sam Altman mentioned in io's launch video is not an in-ear device. The prototype of the device has still not yet been finalized, but it is apparently not a wearable device. The product is still at least a year away from being ready to market.

GROK LEAK HINTS AT SPREADSHEET EDITING

Leaked code inspected by a reverse engineer indicates xAI is adding a Grok file editor with spreadsheet support, signaling a push to embed AI copilots in productivity tools.

REINFORCEMENT LEARNING, EXPLAINED WITH A MINIMUM OF MATH AND JARGON

Reinforcement learning (RL) teaches AI models through trial and error rather than just copying human examples, allowing them to learn from mistakes and stay focused on complex tasks. Although RL has always been a part of training LLMs, companies have recently started using two new approaches to scale up the amount of data available: using AI models to grade another AI model's outputs and creating specialized environments for tasks with verifiably correct outcomes, like coding or math.

NVIDIA TENSOR CORE EVOLUTION: FROM VOLTA TO BLACKWELL

Tensor Cores—the specialized matrix multiplication units inside modern GPUs—are the core innovation propelling recent AI progress. While Moore's Law has stalled due to physical constraints, NVIDIA has doubled Tensor Core throughput with every generation of GPU ("Huang's Law") by adding ultra-low precision data types and introducing asynchronous execution.

EVALUATING LONG‑CONTEXT Q&A SYSTEMS

This post reviews metrics, dataset design, and human or LLM evaluation methods for long‑context QA, outlining challenges like information overload, dispersed evidence, multi‑hop reasoning, and hallucinations.

blog.

REINFORCEMENT LEARNING TEACHERS OF TEST TIME SCALING

Using "teacher" models that explain solutions rather than solve problems from scratch, a small 7B-parameter model outperformed DeepSeek R1's 671B parameters (26.3% vs 18.9% on math benchmarks). Unlike traditional distillation where massive models must first learn to solve problems themselves, these teachers are given both questions and correct answers upfront and trained only to generate clear explanations that help student models understand.

REINFORCEMENT LEARNING SCALING

By randomly pruning a fixed proportion of weights once before training, sparse DRL networks prove more parameter‑efficient and avoid optimization pitfalls.

CUTTING LORA REDUNDANCY

SeLoRA re‑parameterizes LoRA adapters in a sparse spectral subspace, trimming redundant parameters without sacrificing expressiveness and boosting accuracy on commonsense, math, and code generation tasks.

META APPROACHED AI STARTUP RUNWAY ABOUT A TAKEOVER BID BEFORE SCALE DEAL

The list of startups Meta has approached for acquisition continues to grow. It now includes video generation startup Runway. Talks didn't progress far and Meta ultimately pursued a deal with Scale AI instead.

YOU SOUND LIKE CHATGPT

AI changes how we speak by influencing vocabulary and speech patterns, leading to a more uniform language style. Studies show an increase in AI-preferred words and a decline in others, raising concerns about trust and perceived authenticity in communication. The challenge lies in balancing AI integration with preserving linguistic diversity and individual expression.

A DEEP CRITIQUE OF AI 2027'S BAD TIMELINE MODELS

It's really difficult to predict how a technology like AI will turn out - the uncertainty shown in AI 2027 shows how people shouldn't base their futures on prognostication.

TRANSFORMERS BACKEND LANDS IN SGLANG

The new integration lets developers pair Hugging Face's model APIs with SGLang's high‑throughput, low‑latency engine.

hours, you can become an investor at a discounted valuation…

What you need to know 👇

Invest by June 26 before their round closes.

FEDERAL RESERVE REMOVES 'REPUTATIONAL RISK' FROM BANK EXAMINATIONS

The Federal Reserve has eliminated “reputational risk” as a factor in bank exams, aligning with the OCC and FDIC in removing a key regulatory barrier that critics say fueled discriminatory debanking of crypto firms. The change may allow banks to reengage with digital asset companies without fear of regulatory backlash. The move marks a notable shift in U.S. banking supervision amid political scrutiny of Operation Chokepoint 2.0 and growing pressure to clarify crypto access.

CRYPTO MARKETS HOLD AT 'INFLECTION POINT' AFTER BITCOIN'S DIP ON MIDDLE EAST FRICTION

Bitcoin rebounded above $102K after briefly falling below $99K on Middle East tensions. Analysts see markets at an “inflection point,” balancing geopolitical risk and macro uncertainty against growing institutional adoption, including Texas' Bitcoin reserve and new corporate treasury buys. While short-term positioning remains cautious, long-term demand signals resilience, with analysts eyeing July Fed updates and geopolitical developments as key directional catalysts.

A NEW FRAMEWORK FOR TOKEN MARKET TRANSPARENCY

Blockworks launched the Token Transparency Framework, an 18-criterion disclosure standard covering token supply and vesting, insider incentives, market-making and exchange agreements, entity structures, and financials to close critical information gaps in the crypto market. Initial adopters – Jito, Aerodrome, Raydium, Stride, Jupiter, and Morpho – achieved transparency scores above 35 versus a control group scoring 8–13. The open-source framework aggregates materiality-weighted scores across four categories and is intended to streamline due diligence for investors, market makers, and exchanges while complementing SEC initiatives like Safe Harbor 2.0.