Search the archive

A CHINESE AI LAB JUST BUILT A GIANT DATA CENTRE WITH NO NVIDIA INSIDE

Z.ai has started partial operations at a giant data center that runs only on Chinese-made chips.

Get started with Serverless Fine-Tuning now!

KIMI WORK

Kimi Work is an agent that can connect to local files and automate browser work. It is capable of 24/7 automation, running scripts or tasks around the clock quietly in the background. The agent can navigate the internet and execute multi-step web tasks. It can coordinate multiple specialized agents to break down and solve multi-layer tasks and convert insights into professional PowerPoint decks or Excel sheets. Kimi Work is available for both Windows and macOS.

AMD'S HELIOS

AMD has unveiled Helios, its first rack-scale AI system positioned against Nvidia, with Microsoft planning to deploy it in Azure data centers. Meta, OpenAI, and Oracle were also named as early customers ahead of shipments later in 2026.

GOOGLE'S NEW CHIP FOR GEMINI

Google reportedly developed a server chip called Frozen v2 for a potential 2028 release, targeting six to ten times more tokens per unit of power than its existing AI hardware. The project reflected a broader push to reduce inference costs and reliance on Nvidia.

SPARSE BY DESIGN

Kimi K3 activates 16 of 896 experts per token. While the total parameters have grown, the active parameters have barely grown at all over the past three releases. The strategy appears to be to keep per-token compute roughly flat while relentlessly inflating total capacity. At a fixed training compute budget, more experts means lower loss, and the model learns more from the same FLOPs. The open source labs have discovered that the cheapest way to buy intelligence is to spend capacity.

ON KIMI K3: ITS CAPABILITIES AND RELATED DISCONTENTS

Kimi K3 is a very good model with excellent benchmarks. It is the largest (soon-to-be) open model so far, at 2.8T parameters, which explains many of its gains. The model is somewhat distilled, and its performance looks jagged, but it will likely fit well into many workflows. At the current rate of development, China appears capable of releasing a Mythos-level open model by the end of the year.

WHAT LONG-HORIZON AI FAILURES REVEAL ABOUT SAFETY

OpenAI detailed how an internally deployed long-running model exhibited unexpected unsafe behavior that existing evaluations had missed. The company paused access, built new tests, strengthened trajectory-level monitoring, and argued that limited deployment with rollback controls is essential for aligning increasingly autonomous systems.

ONLINE LEARNING FOR COST-EFFICIENT LLM ROUTING

Ramp Router learns provider failure rates through EWMA and latency distributions through Thompson sampling, then chooses the cheapest model and service tier likely to meet each deadline. Ramp reports 30% savings in Ramp Inspect without performance loss.

INTRODUCING COSMOS 3 EDGE

NVIDIA Cosmos 3 Edge is a 4-billion-parameter open world model that helps robots and vision AI agents understand their surroundings, reason in real time, and generate robot actions on edge devices. It is now available on Hugging Face. The model delivers memory-efficient, high-throughput inference across NVIDIA edge computers. It connects understanding, prediction, simulation, and action through a shared world representation. The model can be used as a reasoner or an action generator.

AGENT SWARMS AND THE NEW MODEL ECONOMICS

Every jump in AI capability has raised the level of abstraction at which an engineer works. Agent swarms make the spec the unit of work. Swarms translate intent, but they do this probabilistically, which makes it difficult for them to follow the spec. This post looks at what it takes to make swarms actually follow the spec.

XIAOMI-ROBOTICS-1

Xiaomi-Robotics-1 is a ready-to-use robot foundation model trained on over 100K hours of real-world manipulation trajectories. It combines large-scale embodiment-free pre-training with a modest amount of real-robot data in a post-training stage. It serves as a strong robot foundation model for downstream applications and can learn new tasks with high data efficiency. Footage of robots operating on the model performing household tasks is available in the post.

LANGUAGE MODEL HARNESSES ARE COMPOSITIONAL GENERALIZERS

Scaling data will remain the biggest driver of progress. The machine that we feed that data into and its inductive biases are what will determine the coefficients of that scaling. Better returns on scaling require compositional generalization. The capacity for compositional generalization seems to largely live in harnesses.

Z.AI BUILT A GIGAWATT-SCALE AI DATA CENTER

Z.ai completed a 1-gigawatt data center powered entirely by Chinese-made chips and began partial operations. The facility expanded the computing infrastructure available for training its advanced GLM models.

HOW AI IS SUPERCHARGING DRUG DEVELOPMENT

AI is cutting preclinical costs and timelines in drug development by up to 70%, driving demand for advanced software and models. This could boost new drug program growth by over 10% in three to five years. However, AI has yet to yield an FDA-approved drug, raising concerns about its impact on patient treatment.

WHY YOUR AI BILL WENT UP EVEN THOUGH TOKEN PRICES ARE FALLING

Token prices dropped significantly, yet many enterprises exceed AI budgets. A unit of inference cost $60 per million tokens in 2020, now just pennies. The discrepancy suggests inefficiencies despite lower token costs.

terminate anytime. Launch a cluster on Verda→

ANTHROPIC SET TO END CONWAY TEST AS WIDER ROLLOUT EXPECTED

Anthropic will discontinue its Conway experiment by July 24, prompting users to export data.

MORE AI SPEND WON'T FIX YOUR SUPPLY CHAIN

Sushanth Raman announced the launch of Custom Models for supply chain teams.

WELCOMING TIERZERO TO COGNITION

Cognition has acquired TierZero to enhance software automation in its product, Devin.

X RELAUNCHES A REBUILT ANDROID APP AFTER YEAR-LONG EFFORT

X has relaunched its Android app after nearly a year of rebuilding it from scratch to improve loading, scrolling, notifications, reliability, and future feature development. The effort was described as one of the largest engineering projects in the company's history and is aimed partly at strengthening X in Android-heavy global markets. Some features are still unfinished, including better support for older devices, Spaces, video editing, react-with-video, cashtags, and custom timelines.

APPLE BRIEFLY BECAME THE WORLD'S MOST VALUABLE COMPANY, TOPPLING NVIDIA

Apple briefly overtook Nvidia as the world's most valuable company after investors sold off AI-related semiconductor stocks, reflecting growing skepticism about the long-term returns of massive AI infrastructure spending. Unlike many of its rivals, Apple has spent relatively little on building AI infrastructure and instead benefits from its installed base of 2.5 billion devices. As a result, it is increasingly seen as well positioned to monetize AI features, especially after the positive reception to its revamped Siri.

AGENTIO'S AI-POWERED CREATOR ADS ARE COMING TO FACEBOOK AND INSTAGRAM

Agentio, an AI-powered creator advertising platform previously focused on YouTube, has partnered with Meta to bring its tools to Facebook and Instagram. Brands can now access AI-driven insights, creator matching, and partnership ads across both platforms, with the option to run cross-platform campaigns spanning Meta and Google properties. Beta testing showed Agentio's Meta partnership ads achieved 81% higher ROAS and 89% higher click-through rates compared to brands' non-Agentio creator ads.

PRODUCT QUALITY: A SHARED COMMITMENT TO CRAFT IN THE WAKE OF AI

As AI collapses the gap between "nothing" and "adequate," genuine craft becomes the only remaining differentiator between products. Craft is a shared discipline across Utility, Usability, and Feel—owned collectively by engineers, PMs, designers, and leaders, not design alone. "Would I recommend this to someone I care about?" should be the standard for judging whether a product has truly earned people's affection, not just their approval.

WHY THE FUTURE OF DESIGN MUST BEGIN WITH UNDERSTANDING PEOPLE

User experience has always extended beyond visible interfaces, shaped instead by trust, psychology, and human context. Drawing on architecture, emergency medical care, and design experience, the author contends AI heightens designers' responsibility to build informed—not blind—trust, making AI literacy an essential design skill. Design's future depends on understanding people rather than mastering tools, since technology changes but human needs remain constant.

ACCESSIBILITY ISN'T A CHECKLIST. IT'S A DESIGN DECISION YOU'RE AVOIDING

A UX designer recounts watching a blind user struggle through an 11-minute flight booking process that should have taken 90 seconds, revealing how years of seemingly reasonable decisions had quietly excluded users with disabilities. The experience showed that accessibility isn't primarily a technical challenge or a compliance checklist—it requires making it a core product priority, testing with real assistive technology users, and treating accessibility as part of the definition of "done" rather than something to address later.

AI SKILLS FOR PRODUCT DESIGNERS

Whether you're directing your AI or working alongside it, Layers walks you through all seven layers of product design, so the decisions underneath the screen actually get made.

DATATYPE

Datatype is an OpenType variable font that turns simple text expressions into inline charts. No JavaScript, no images, and no rendering library.

REMIX HIGH-PERFORMING ADS FOR YOUR BRAND

Browse a wall of proven ad creatives from top brands, drop in your brand, and get an on-brand video or static ad in minutes.

AGAINST DESIGN SYSTEM FEDERATION

Federated design system governance—used by just 13% of teams—is losing favor due to poor contribution rates, diffused accountability, and inflated cost-saving expectations. Layoffs and AI-driven productivity narratives have fueled interest in federation, but it typically damages system quality, user experience, and team morale rather than fixing organizational problems. Federation works best as one aspect within a hybrid model, not as a standalone governance approach, since it demands more organizational maturity than most companies achieve.

THE “MOCKED, THEN LOVED” AIRPODS EFFECT

Apple's AirPods and Apple Watch both overcame early ridicule by proving genuinely useful in everyday life, but the rumored camera-equipped AirPods face a different challenge: concerns about privacy and social acceptance rather than just unconventional design. While Apple has a strong track record of turning skepticism into success, the new AirPods will need to demonstrate clear, hands-free AI benefits while convincing both wearers and those around them that the cameras can be trusted.

ASK YOURSELF WHERE YOU CAN ADD VALUE

Designers should stop undervaluing in-house work and instead showcase the business problems they solved, the constraints they worked within, and the measurable results they achieved. Rather than waiting to build a perfect portfolio, improve existing projects incrementally with stronger storytelling, updated visuals, and small enhancements, making steady progress in short bursts that fit around a busy schedule.

HOW AI HELPED US MAKE SENSE OF HUNDREDS OF COMPONENTS

A UK higher education institution had years of organically grown web components with no clear inventory, so a team built an AI-assisted audit tool through rapid, conversational iteration rather than a fixed spec.

THIS MAC APP CAN TURN EVERY WINDOW INTO A CRT FEVER DREAM OR A NOSTALGIC GAME BOY

Glaze is a Mac app that transforms your entire display with over 50 real-time visual effects—from CRT monitors to VHS tapes—while also offering productivity and accessibility modes.

BRANDED SEARCH IS BECOMING A LESS RELIABLE PROXY FOR BRAND DEMAND

Branded search has long been a proxy for brand demand, but it's becoming less reliable. Data shows branded queries dropping even as demand holds steady, pointing to a measurement gap, not a demand decline. AI search is shifting decisions upstream, answering questions before users search for brands. The risk is misreading weaker search as weaker brands and underinvesting in brand building. Branded search still matters, but it should be one signal, not the source of truth.

DOES TOPICAL AUTHORITY MATTER IN AI SEARCH?

Topical authority still matters in AI search, with data from Semrush's analysis of 50,000+ brands showing AI answers develop category owners over time. In ChatGPT's June 2026 data, only 15.2% of categories had a clear owner, leaving 89.3% of AI search demand in categories without one. Category owners also showed stronger brand search volume, organic traffic, and authority. Build depth in core topics rather than chase broad content coverage.

WHAT EVERYONE GETS WRONG ABOUT CATEGORY CREATION

Category creation only works when it reflects a real job to be done instead of a new marketing label. Companies create confusion when they invent platform names that don't match a clear customer responsibility or workflow. It's better to create a new job to be done before naming a category, sell products as a suite when buyers purchase them separately, or lead with the product the market already understands and expand from there.

THOUGHTLESS FEATURES NEVER ADD UP TO A THOUGHTFUL PRODUCT

Teams ship “ham biscuit” features fast, solving surface problems but missing real user needs. Idea generators stay detached, copying competitors or AI outputs, while execution follows without question. The result is bloated products that hurt clarity and sales. Better products come from defining real workflows, not stacking features. Focus on the job to be done, validate outcomes, and design end-to-end experiences instead of isolated fixes.

HOW TO CHECK CHROME'S ACCESSIBILITY TREE

AI agents don't view websites like humans do. They rely on the Accessibility Tree, a machine-readable version of your page structure. Check yours in Chrome DevTools to see how agents interpret your content. Clean up headings, labels, links, and alt text to make your site easier for AI systems to understand and surface.

SUBJECTLINE.COM

This is a free email subject line tester that scores over 15M subject lines and helps improve email performance. Free accounts provide subject line history, score tracking, and AI-powered rewrites. Pro accounts ($9.99/year) add open rate predictions, industry benchmarks, and project organization tools.

RCS OR SMS? IT DEPENDS ON THE TASK

RCS now reaches 96% of US devices and offers rich media, interactivity, and branding, but costs at least twice as much as SMS. Like HTML email, it adds capability, not necessity. Use RCS when visuals or actions help complete a task like browsing products or resolving issues. Stick with SMS for simple, urgent updates like order confirmations or codes.

AI IS WORKING OVERTIME. SO AM I

AI users save an average of 5.1 hours a week but lose 3.3 hours learning or troubleshooting the same tools. 77% report a heavier workload, while 23% report AI-related anxiety. The shortfall comes from generation time being mistaken for completion time, since a report that takes one minute to draft still needs its research checked and its logic fixed before it is usable. The technology accelerates the first draft while leaving the harder work, review and accountability, exactly where it was.

THE BIGGEST TECHNICAL SEO TIME-WASTERS TO AVOID

Chasing minor SEO fixes like redirects, crawl budget, 404s, or tool flags wastes time when they don't impact rankings or user experience.

MY GOAL IS TO NEVER APPLY FOR A MARKETING JOB EVER AGAIN

Build a personal brand before you need it by staying visible and optimizing LinkedIn.

TELL YOUR ENGINEER TO CHANGE "LOADING..." TO "THINKING..."

Products that hide their "thinking" step read as less warm to users.

8 STRATEGY CONCEPTS EVERY PM SHOULD KNOW

Strong product strategies combine several sources of competitive advantage. Those advantages must be chosen carefully and revisited as the market changes.

IS "JOBS" A FOUR-LETTER WORD?

AI is changing Silicon Valley's definition of ambition from building large teams to needing fewer people. This shift could make hiring look inefficient and layoffs look impressive.

YOUR AI SHOULD KEEP A DIARY

An AI activity log makes invisible work easier to review and improve. It helps users delegate more effectively, preserve context, and build stronger AI habits.