Search the archive

EVERY FRAME PERFECT

Every frame of an interface should feel intentional.

ANTHROPIC SHUTS DOWN FABLE, MYTHOS MODELS FOLLOWING TRUMP ADMIN DIRECTIVE

Anthropic has shut off access to its Mythos 5 and Fable 5 models. The US Commerce Department has issued a directive that subjects the new models to export controls restricting their use anywhere outside of the US. Anthropic says the only way to ensure compliance is to completely cut access until the issue is resolved. Access to other Anthropic models is not affected.

ELON MUSK BECOMES THE WORLD'S FIRST TRILLIONAIRE

Elon Musk became the world's first trillionaire on Friday after shares of SpaceX rose 20% from their initial public offering price of $135. Musk was already the world's richest person, claiming the title from Jeff Bezos in 2021 with a net worth of over $185 billion. His worth is now equivalent to more than 3% of the US gross domestic product. It is five million times as large as that of the typical US family.

NUCLEAR CLOCKS TICK FOR THE FIRST TIME

Chinese and European research teams working independently have both achieved building a working nuclear clock. Both exploit the nucleus of a thorium-229 atom to keep time with extraordinary precision. Building a working nuclear clock has so far proven to be challenging, as the required laser light sits in a part of the spectrum that is technically challenging to generate and control. The Chinese team used a more powerful laser, while the European team worked with a crystal containing a higher concentration of the thorium nuclei.

NEW CRISPR TECHNIQUE SELECTIVELY SHREDS CANCER CELLS, INCLUDING “UNDRUGGABLE” CANCERS

Researchers have created a new CRISPR-based approach that can selectively destroy cells carrying a mutation in a tumor suppressor found in nearly half of cancers and up to 70% to 90% of cases of some of the most difficult-to-treat cancers. The new approach harnesses CRISPR's ability to find cells with specific mutations and uses its cutting ability to selectively destroy those cells. Making treatments for different cancers using this technique is faster than making a small molecule drug or antibody therapy. Delivery remains a challenge, but a combination of therapies may prove useful for some cancers.

starting June 24. Save your spot

PONYTAIL

Ponytail is an AI senior developer. It produces efficient code at low cost and at fast speeds. Ponytail works with every model. Examples and benchmark comparisons are available in the repository.

THE LOG IS THE AGENT

While the model, runtime, and loop matter, those things aren't the agent. The agent is the history of events that make up its data. This log allows agents to be reconstructed correctly and resumed. The concept of agents as logs unlocks a whole set of capabilities that make even advanced agent use cases easier to reason about and build into everyday applications.

BONUS, FULLY REMOTE)

We're hiring a senior PMM to own product marketing at TLDR. You'll define our positioning, build out sales enablement, and lead every launch. Learn more [12].

APPLE'S NEW SIRI IS JUST GOOD ENOUGH TO EASE ITS AI CRISIS

Apple's Siri now works the way it's supposed to. The latest update doesn't do anything revolutionary or innovative, but it shows how Apple has finally entered the modern AI market. The new Siri is roughly competitive with where the leading chatbots were about six months ago. Apple's preinstalled functionality is now competent enough that many users won't need competing AI services.

META'S MONTHS-OLD AI UNIT IS A SOUL-CRUSHING GULAG, SAY THE ENGINEERS STUCK INSIDE IT

Meta employees don't think it's a happy place to work at, especially given the seemingly endless layoffs the company has executed over the last few years. The company's Applied AI team is reportedly on the verge of revolt. One of the group's live-streamed, employee-only presentations was interrupted earlier this week with an expletive-laden meltdown targeting a senior Meta AI executive. Many employees in the group say they were forced into the group with no real choice and that the work inside the unit was soul-crushing.

scale. Dive deeper into the shift

HAS AI ALREADY KILLED HOW-TO NONFICTION? SALES TRENDS, MY PERSONAL DATA, AND WHAT IT MIGHT MEAN FOR THE FUTURE

The market for information is collapsing into the chatbot.

INSIDE THE ROOM WHERE AMERICA'S BRIGHTEST GAME OUT HOW TO AVOID AN AI APOCALYPSE

Political polarization makes ambitious changes difficult, but inaction would be far more costly.

LEAVING MOZILLA

JR Conlin is leaving Mozilla on July 21 after more than 15 years.

Explore SPAL & SNK

PROP-FOR-THAT

prop-for-that exposes what JavaScript knows, but CSS can't see as live CSS custom properties.

SPACEX'S PRESIDENT IS FLOATING A TESLA MERGER AS THE COMPANY BEGINS TRADING

SpaceX President Gwynne Shotwell declined to rule out a future merger with Tesla while tempering expectations about timing.

SOFTWARE IS NOT A SINGLE-PLAYER GAME

Building dependable software that lasts and grows will always remain a multiplayer game.

Use any model with Strands Agents (6,500 GitHub stars).

Your harness runs any model on any cloud. You get context management, execution limits, and observability before you write a line of config.

ANTHROPIC SUSPENDED ACCESS TO FABLE 5 AND MYTHOS 5

Anthropic said it disabled Fable 5 and Mythos 5 for all users after receiving a US government export-control directive tied to national security concerns and reported jailbreak risks.

AMAZON CEO'S TALKS WITH US OFFICIALS TRIGGERED CRACKDOWN ON ANTHROPIC MODELS

Conversations between Amazon chief executive Andy Jassy and US officials prompted the Trump administration to halt all foreign use of Anthropic's most capable AI models. Researchers at Amazon had used a series of prompts to get Anthropic's Fable 5 model to provide them with information that could be used to aid cyberattacks. White House officials asked Anthropic to fix the vulnerabilities or take down the model. Anthropic has shut down access to its Mythos and Fable models to comply, but says that the vulnerabilities flagged by Amazon are relatively basic and that other publicly available models are also capable of discovering them.

GLM-5.2

GLM-5.2, a new flagship model from Z.ai, is now available to all GLM Coding Plan users. It delivers powerful coding capabilities, usable 1M-context support, and continued strengths in long-horizon tasks. API and chatbot services will launch next week. The model will be open-sourced under the MIT License.

GOOGLE IS WORKING ON SKILLS MARKETPLACE FOR GEMINI BUSINESS

Google is integrating its products under Gemini Enterprise with a new "Skills Marketplace" tab for pre-defined, Google-optimized skills. The marketplace, aimed at helping teams develop dashboards and reporting tools without long engineering delays, includes a Skills Management UI, a Skills Builder, and the Marketplace itself.

INFERENCE COST AT SCALE WITH NAPKIN MATH

You need the following information to work out the dollar price-per-user: GPU hardware specs, context length, active parameter count of the model, and product-specific factors. The specifics of the model architecture matter surprisingly little, unless it's something entirely different, like diffusion. This post shows how to work out the math on paper. The exercise should reveal how various optimizations in inference engines help SaaS products remain profitable.

TODAY'S FRONTIER AI COMPANIES WILL NEVER EXCEED THE AI CAPABILITY FRONTIER AGAIN

Networks of smaller AI models are outperforming every frontier AI system on speed, accuracy, and cost. Everyone in the 1960s was wrong about the mainframe computer, and everyone is now wrong about centralized AI. The future is a network of neural networks.

THE PHYSICS OF A FABLE

Rafa Schwinger reverse-engineers Claude Mythos and Fable by arguing the moat is not architecture but the environment foundry, with capability decomposing as base foundation times gradeable signal extracted on top, and verifiable reward becoming the scarce decisive input now that text and raw compute no longer are. The recipe stacks dense pretraining, GRPO-style verifier RL where reward-hacking soundness is the actual binding constraint, long-horizon process rewards with learned context-folding that beats million-token windows at 32K active, plus best-of-N test-time compute exposed as an effort dial.

happen next. Get your copy

KIMI K2.7 CODE (HUGGING FACE REPO)

Kimi K2.7 Code is a coding-focused agentic model that has stronger end-to-end task completion across complex software engineering workflows and improved token efficiency compared to Kimi K2.6. The Mixture-of-Experts model has 1 trillion total parameters. It can be accessed on Moonshot's OpenAI/Anthropic compatible API. The model works best with Kimi Code CLI as its agent framework.

INTRODUCING THE OPEN KNOWLEDGE FORMAT

The Open Knowledge Format is an open specification that formalizes the LLM-wiki pattern into a portable, interoperable format. It is vendor-neutral and agent- and human-friendly. The standard can represent the metadata, context, and curated knowledge that modern AI systems need. The specification uses familiar patterns with no complex compression scheme, new runtime, or required SDK.

OLMO-EVAL: AN EVALUATION WORKBENCH FOR THE MODEL DEVELOPMENT LOOP

olmo-eval is a new evaluation workbench designed for iterative LLM development. Enhancing the OLMES standard, it streamlines adding benchmarks, supports agentic and multi-turn evaluations, and facilitates analysis by comparing changes across model checkpoints. Unlike Harbor, olmo-eval offers flexibility, minimizing resource use and focusing on development rather than public benchmarking.

MINIMAX SPARSE ATTENTION FOR MILLION-TOKEN CONTEXTS

MiniMax Sparse Attention is a sparse attention architecture that uses group-specific Top-k block selection to scale long-context inference while preserving model quality. On a 109B multimodal model, it matched GQA performance while cutting attention compute by ~30x at 1M tokens.

COUNT ANYTHING

Object counting is still fragmented across domain-specific data sets and task formulations. Existing counting models are often tailored to specific scenarios and struggle to generalize across categories, visual domains, object scales, and density distributions. This paper presents a generalist model for text-guided object counting that achieves strong accuracy and multi-domain generalization.

BONUS, FULLY REMOTE)

We're hiring a senior PMM to own product marketing at TLDR. You'll define our positioning, build out sales enablement, and lead every launch. Learn more [20].

THE ORACLE AND THE FIRM

OpenAI and Anthropic have decided on different approaches to context management. OpenAI uses compaction, compressing everything and retaining only the relevant information. This results in one long thread that maintains a lot of coherence. Anthropic splits context windows across various agents, where each agent executes on the sub-problem within its own context window. Sub-agents do a large amount of work, then pass back only the relevant information to the parent agent. Anthropic's approach can result in sub-agents doing duplicate work, forgetting information, and generally wasting more tokens.

WHY APPLE BUILT A THIRD-PARTY AI SYSTEM FOR SIRI AND THEN REFUSED TO SHOW IT AT WWDC

The iOS 27 beta contains an Extensions system for third-party AI. The system includes a settings panel and a dedicated App Store section. Both have been built, but are toggled off in the backend. Apple was in discussions with major AI providers about granting entitlements for the framework, but it appears the company has decided against announcing the feature for now.

$10,000,000 ON THE LINE: HOW WE MEASURE DEVIN'S ENGINEERING OUTPUT

Devin's engineering system guarantees more output than cost, with $10M pledged per customer. The system's effectiveness is validated using independent data. This bold claim aims to bolster confidence in their engineering productivity.

decide whether AI pays off. WATCH ON DEMAND →

NVIDIA BLACKWELL LEADS ON FIRST AGENTIC AI INFRASTRUCTURE BENCHMARK

NVIDIA's Blackwell Ultra NVL72 platform leads the AgentPerf benchmark by delivering 20x more agent throughput per megawatt than NVIDIA Hopper.

RAMP SWE-BENCH: A PRIVATE, PRODUCTION-GROUNDED CODING BENCHMARK

Ramp released its own private SWE-Bench built from real engineering problems faced at Ramp, giving the team a way to evaluate coding models inside its actual financial software ecosystem.

A FRONTIER WITHOUT AN ECOSYSTEM IS NOT STABLE

Microsoft CEO Satya Nadella argues every company needs to build both human capital and token capital.

DEEPSEEK'S $10 TRILLION LONG-TERM STRATEGY

This thread explores a theory that DeepSeek's focus on open-source models, research sharing, and infrastructure development is part of a broader strategy aimed at becoming foundational AI infrastructure rather than competing directly on consumer products.

ICON COMPOSER 2 AND SF SYMBOLS 8 NOW AVAILABLE AS BETAS

Apple has released beta versions of SF Symbols 8 and Icon Composer 2 for developers and designers following WWDC26. SF Symbols 8 adds new symbols across Apple's latest operating systems, bringing the total library to more than 7,000 symbols with support for animations, effects, and variable rendering. Icon Composer 2 introduces Liquid Glass design tools such as Refraction for realistic light distortion, improved specular highlights for better layer definition, and an Extended Preview mode that lets designers see how app icons will appear on both current and previous operating systems.

CAPTURE WEBPAGES AS EDITABLE FIGMA LAYERS WITH THE CHROME EXTENSION

Figma released a Chrome extension that captures webpages as editable layers, allowing users to copy full pages or selected elements and paste them directly into Figma. The feature enables designers to reference and modify web content without requiring coding skills. This beta feature is currently available only on paid plans, with design system integration coming soon.