Search the archive

EVERYTHING HACKABLE WILL GET HACKED

AI models have become much more capable of performing cybersecurity work over the past year. Defenders have an advantage right now because they can use stronger models, but the gap between open-weight models and closed models will soon close. Everyone can start improving their defensive posture with currently available frontier models. Continuously improving how you find and fix vulnerabilities should be a top priority, and the practice should continue as models improve.

SOFTWARE ENGINEERING AT A PROPRIETARY TRADING COMPANY: OPTIVER

Trading companies build bespoke hardware stacks and have larger platform engineering teams than most workplaces. For software engineers, it's a lucrative niche in terms of compensation, full-stack work, and engineering challenges. Software engineers tend to know little about trading companies. This post covers in depth how engineering works in a modern trading business.

HIS START-UP'S GOAL: AI THAT IS TRAINABLE AND NOT CONTROLLED BY A BIG COMPANY

Igor Babuschkin, who founded xAI with Elon Musk, left a year ago to build a new startup, River AI. The startup wants to create a new type of home computer that allows people to own and use AI without oversight from anyone else, including River AI. It plans to share its AI technologies as open source software. Babuschkin envisions future AI systems as trained to benefit individuals rather than centralized and controlled by one big corporation.

ELON MUSK'S NEW TEXAS FACTORY WILL BE THE LARGEST BUILDING ON EARTH, IF ALL GOES ACCORDING TO PLAN

Elon Musk's Terafab in Grimes County, Texas, is slated to top 100 million square feet. The factory will function as a joint supply chain for SpaceX and Tesla. It will combine logic, memory, and advanced packaging under one roof. Construction is projected to cost $16.8 billion. Eight tax incentive applications have been filed on behalf of the project, asking for tax breaks so the project has a more favorable after-tax return.

your PRs. Try Gitar Free

COMPRESSION IS PREDICTION

In compression, entropy measures how small we can shrink things, but in language modeling, it's used to make models better at prediction.

NVIDIA'S RISKY BUSINESS

Nvidia's ability to compete through novel funding mechanisms draws on things like insurance floats and pension funds and other long-run liabilities, which means the risk is considerably higher.

SCALING AN AGENTIC CODING SDK: WHAT CONCURRENCY ACTUALLY COSTS

Agent concurrency is a large number of partially trusted computers operating at once.

MAKE EVERYTHING LOOK LIKE A SOFTWARE PROJECT

There's now a strong incentive to make every project look like a software project.

ROADMAP DECISIONS RATHER THAN DATES

Some timeline tradeoffs are constrained by approvals, cross-team and cross-functional handoffs, or missing decisions, but almost none should be constrained purely on time.

MODULAR 26.5: MOJO 1.0 IS HERE!

Mojo has grown into a general-purpose language with a vibrant developer community.

MARK ZUCKERBERG LAYS OUT NEW AI VISION IN 6,500-WORD ESSAY

Mark Zuckerberg recently released a wide-ranging essay that outlined a new course of action to spread the wealth and opportunity from AI to users around the world, and to residents of the communities that host the data centers powering it. Zuckerberg's plans include releasing more open-weight models and establishing a $1 billion fund to invest in the communities where Meta hosts data centers. He believes that Meta's approach to developing powerful AI and distributing it as widely as possible is the least likely path to disastrous outcomes for humanity. The commitment marks a significant development in the debate around whether advanced AI is too dangerous to create on an open-source basis.

GOOGLE'S CLASSIC SEARCH BUTTON IS GONE IN A NEW AI-FIRST HOMEPAGE

Google is testing a version of its homepage where the search button is replaced with an AI Mode button. The new interface allows users to ask about files and use the Brainstorm function without having to log in. Logged-in users are also able to generate images. Ask about files allows users to attach a file before submitting a question, and Brainstorm helps users explore ideas.

DYNA-2 PROVES SCALING LAWS FOR ROBOTICS: 1 MILLION HOURS OF HUMAN VIDEO UNLOCKS ZERO-SHOT DEXTERITY

Dyna-2 is a world-action model (WAM) pre-trained on over one million hours of human video data. Its existence proves that scaling human video data predictably improves zero-shot performance on unseen robot hardware. Dyna-2 achieved an 87% pass rate in real-world zero-shot deployments at customer sites, vastly outperforming the 46% pass rate of its VLA predecessor, Dyna-1. The model uses a new one-step video generation distillation pipeline that drops inference latency by two orders of magnitude for downstream planning.

LEARNING MORE ABOUT CLAUDE'S MATHEMATICAL CAPABILITIES

Claude was recently challenged with one of the most famous unsolved problems in mathematics: the Riemann hypothesis. While Claude didn't succeed, it unexpectedly made strides on a related problem. It improved on a longstanding lower bound for the fraction of zeros of the Riemann zeta function that satisfy the Riemann hypothesis. The result shows that AI models like Claude can extend the impact and reach of mathematicians' ideas in new and sometimes surprising ways.

NO, LOCAL MODELS WILL NOT WIN

Local models are never going to be as powerful as datacenter models. All of the current frontier models are far too big to run on anything but a full GPU cluster. Smaller models are getting more intelligent, but datacenter models will always be the strongest available. When given a choice, people are going to pick the strongest available model in their price range, the one that frustrates them least, which is always going to be the bigger, more powerful one.

WHAT'S THE BEST PROGRAMMING LANGUAGE FOR CODING AGENTS?

Anyone who wants to make a strong claim about which language or classes of languages are particularly good with LLMs would need to run quite a few different evaluations. Failures with languages are generally something idiosyncratic, where it's not always obvious how much the issue generalizes across tasks or setups. There's no way to look at the score on one or even several evaluations and draw a conclusion about programming in general. At this point, it would be a mistake to draw a strong conclusion about any particular language.

SEE HOW A TESLA-SPACEX MERGER GIVES MUSK A SHORTCUT TO HIS $1 TRILLION PAYDAY

The conditions for Elon Musk to receive his $1 trillion pay package at Tesla could easily be achieved if the company were acquired. There is a clause in the agreement that declares half the targets as good as accomplished if Tesla is acquired or otherwise taken over. This leaves only market-value targets to determine how many shares Musk gets. Musk has considerable say over any offer SpaceX makes for Tesla. Even a stratospheric bid would keep Musk in firm control of the combined company.

NVIDIA LINES UP $500 BILLION IN FINANCING AS CEO JENSEN HUANG TELLS CNBC HIS CHIPS ARE ‘INVESTABLE ASSET'

Nvidia has partnered with six large asset managers on a $500 billion financing push designed to treat compute infrastructure much like commercial real estate, toll roads, or other assets to borrow against. The effort will mobilize third-party capital for hyperscalers, frontier AI labs, and enterprises to build out data centers and acquire Nvidia hardware. It will help Nvidia's customers secure financing without tapping their own balance sheets. Nvidia believes lenders can reliably underwrite compute as a revenue-generating asset with an extended life as its hardware is broadly adopted and transferable across customers.

data and deploy dedicated instances all in one place. Start now

EVERYTHING I LEARNED SHIPPING DEVICE BOUND SESSION CREDENTIALS

Device Bound Session Credentials bind sessions to a private key a browser generates in hardware and cannot export.

FOLLOWING EPIC LOSS, GOOGLE HAS STARTED HOSTING RIVAL APP STORES IN THE PLAY STORE

Aptoide, a game-focused store that has been available as a sideloaded client since the early days of Android, is the first third-party app store to arrive in Google Play.

INTRODUCING MUSE GLIMMER

Muse Glimmer is a 30B open weights model optimized for end-to-end agentic task completion, reliable tool use, and multi-step reasoning.

APPLE IS WORKING ON A WAY TO AUTHENTICATE THAT A PHOTO CAME FROM AN IPHONE CAMERA

Apple Reference Image is designed to authenticate the source of photos using unique data tied to the iPhone camera hardware that captured them.

A RESEARCHER BOUGHT NOREPLY.NET. COMPANIES STARTED SENDING HIM SECRETS

Cory Solovewicz created an accidental honeypot when he purchased noreply[.]us and noreply[.]net in 2020 and 2024, respectively.

COMPUTE IS REVENUE. REVENUE IS COLLATERAL

Nvidia is attempting to turn a handful of bespoke transactions into a financial production line.

Take a closer look

APPLE EXPLORES SHAKING UP ITS SMARTWATCH FOR A NEW ERA

The Apple Watch continues to attract new buyers and remains the top-selling smartwatch. However, the company's industrial design team has been exploring a broad rethink of Apple's smartwatches and is evaluating several different directions. Apple needs to adapt to the rise of screenless fitness bands and smart rings, which are cheaper, lighter, and more fun and convenient to use for many people. The company aims to inject new life into its Wearables, Home, and Accessories segment, which has struggled to generate meaningful growth in recent years.

NOW WE HAVE A TIMELINE OF THE OPENAI ACCIDENTAL ATTACK AGAINST HUGGING FACE

OpenAI recently gave a presentation about the Hugging Face incident. The talk is short and information dense and provides full details of what happened and how things played out inside OpenAI. This post presents the timeline of events and details how the hack occurred. A video of the presentation is available in the post.

THE FIRST SELF-DRIVING VEHICLE ON MARS HAS PROVEN TO BE A SMASHING SUCCESS

NASA's Perseverance Mars rover will set a record for the most distance driven by any vehicle on another world sometime next week. The rover's auto navigation system has enabled about 90% of the distance driven by the vehicle to be autonomous. Not having to wait for navigation commands from drivers on Earth has allowed the rover to maximize the amount of scientific return. The vehicle will likely be able to keep driving itself across Mars for many years to come.

CHINA-FREE BATTERIES MADE FROM SALT ARE FINALLY HERE

Peak Energy, a US-based startup, has raised tens of millions of dollars to build a factory for its sodium-ion battery packs. It already has more than $1.1 billion in announced deals with customers. Sodium batteries have less-intensive cooling needs than lithium-ion packs, and they are less likely to catch fire. Another US-based startup, Inlyte Energy, is working on another type of sodium-based battery that can't catch fire at all.

AGENTIC CODE QUALITY

Software quality now depends on the constraints set around the agents that wrote it. Constraints define what a system is allowed to do. Setting and maintaining these constraints allows developers to build loops that reliably deliver high-quality production software. Agents can propose anything - constraints decide whether a proposal is safe enough, correct, scoped, and useful.

GOOGLE'S WESTINGHOUSE BET

Google's recent reshuffling of its AI operations has many thinking that the company has failed in the AI race. However, the company may just be refocusing its resources on its strengths. Google Cloud may be a much larger economic opportunity than pursuing its rivals in the frontier AI race. The company's enormous distribution advantage means it doesn't have to compete in the expensive and unprofitable frontier race.

AMAZON CRACKS DOWN ON 'CPU WASTE' AMONG ENGINEERS AS AGENTIC AI CRUNCH INTENSIFIES

AWS is cracking down on internal use of EC2 instances among its engineers. Engineers have been told to reduce CPU waste to ensure the company has enough CPU capacity to meet customer demand. The increased demand for CPU is due to the rise of AI agentic workloads. Other cloud companies are also seeing high demand for CPUs.

CHANGING DEVTOOLS IS CHEAP. OWNING THEM ISN'T

The future of software will be powered by high-quality, well-documented building blocks that users can build specialized artifacts on top of. Agents are good at composing high-quality components. Maintainers can provide these components alongside a focus application, and users can choose how to modify their experiences themselves. This allows software to be personalized while still being developed by maintainers that care.

WHY YOU MIGHT WANT TO BUILD YOUR WEBAPP IN CANVAS INSTEAD OF HTML

Canvas is a low-level rendering tool that gives users more control than HTML.

INTERVIEWING ENGINEERS IN THE AI ERA: LESSONS FROM A YEAR OF REBUILDING

Coinbase has rebuilt its engineering interview loop to test how candidates direct AI, evaluate its output, and apply judgment where models fall short.

CHATGPT STARTS BLOCKING DIRECT REQUESTS TO COPY AN AUTHOR'S STYLE

ChatGPT will now instead offer a response that draws on the broad qualities of those authors while remaining distinct in its own voice.

REVISION PROMPTING IMPROVES INDUSTRIAL LLM PROCESSES

Revision Prompting supplies LLMs with the original input, the original output, and the changes to the input, and instructs it to produce a patch that updates the original output accordingly.

HOW I REBUILT MY DEVELOPMENT WORKSTATION FOR AGENTIC WORK

This post describes a workstation that is terminal-native, agentic-first, and built around composable tools.

OPENAI'S NEW DEVICE WILL BE HOCKEY PUCK-SIZED AND COST OVER $300

OpenAI's upcoming hardware device will be essentially a displayless smart speaker with moving parts to help give it a personality. The product is slated for release in 2027 for a cost of more than $300. OpenAI is positioning the device as an AI-first computer that can help users get things done. Its design will apparently set it apart from other current smart speakers.

TESLA, SPACEX CONFIRM ‘TERAFAB' CHIP FAB SITE — $16.8B FIRST PHASE

Tesla and SpaceX will build their Terafab megafactory in Grimes County, Texas. Terafab will combine logic, memory, packaging, and testing under one roof. The initial phase will cost roughly $16.8 billion, with the finished site spanning more than 100 million square feet. SpaceX is aiming to produce over a terawatt of compute per year at the site.

THIS AI JUST CREATED VIRUSES NOT FOUND IN NATURE

Researchers have used AI to create new kinds of viruses for the first time. They taught AI to recognize patterns of DNA structure in nature and then use that data to write recipes for entirely new viruses. The researchers followed those recipes to create DNA molecules, which they inserted into bacteria. The viruses were able to infect other bacteria, demonstrating that they were viable.

NUCLEAR STARTUP OKLO SPLITS ITS FIRST ATOMS IN TEST REACTOR

Oklo's Groves Isotope Test Reactor has sustained a chain reaction. The reactor is a demonstration version of the technology Oklo plans to use for its new medical isotope business. The company also plans to expand into recycling nuclear waste, fabricating its own fuel, and producing medical isotopes. Oklo is preparing a site near the Idaho National Laboratory for its first small modular reactor but does not yet have permission to work on the reactor itself.