Search the archive

PROGRESS

Etched's new architecture enables AI chips to run math blocks at half the usual voltage, boosting FLOPs density and tackling thermal issues.

AMD AND CEREBRAS LAUNCH AI INFERENCE SOLUTION

AMD and Cerebras have unveiled an AI inference solution that combines AMD Helios with Cerebras Wafer-Scale Engine, enhancing low-latency AI workloads.

READ THE DEEP DIVE→

START FREE. $100 IN CREDITS →

OPENAI PRESENCE

OpenAI Presence is an enterprise product for deploying controlled AI agents across customer support and internal operations. It combines model reasoning with permissions, policies, evaluations, escalation rules, and tools for improving agents after deployment.

AMD AND ANTHROPIC SIGN MAJOR CHIPS-AND-INVESTMENT DEAL

AMD and Anthropic have signed a deal for tens of billions of dollars' worth of AI servers. Anthropic will purchase up to 2 gigawatts of AMD's Instinct MI450 chips starting in the first half of next year. AMD will invest up to $5 billion in Anthropic as certain deployment milestones are met. The companies are working together to identify data centers for the chips.

ARE AI LABS PELICANMAXXING?

Simon Willison's 'generate an SVG of a pelican riding a bicycle' benchmark has become one of the most famous informal benchmarks in AI. It is famous enough that there is a lot of discussion about its usefulness and whether AI labs might be making their models specifically better at the task. Testing has found that there is little evidence that AI labs are doing this. At least, not in a plainly obvious manner.

OPENAI'S ACCIDENTAL CYBERATTACK AGAINST HUGGING FACE IS SCIENCE FICTION THAT HAPPENED

OpenAI was running a cybersecurity test against an unreleased model with the model's guardrail features turned off. The model broke out of OpenAI's sandbox and found exploits to break into Hugging Face so it could cheat on the test by stealing the answers. This scenario makes the strongest case yet for how the imbalance of model availability is hurting our ability to secure our software. The frontier models we have access to are increasingly being constrained in how much they can help us protect our software, and this is heavily influenced by the US government's ongoing threat of export controls.

NOBODY KNOWS WHAT A USED GPU CLUSTER IS WORTH

GPU collateral has three values, and the market is currently pricing only one of them. Face value is the purpose price minus straight-line depreciation. Liquidation value is what a buyer pays in distress. Going-certain value is what the cluster is worth as a working asset to the next tenant. The spread between face value and going-certain value is the entire risk that nobody has hedged.

REGISTER NOW

CURSOR ROUTER

Cursor Router is an intelligent model router that selects the right model for the task at hand. It delivers frontier-quality results at 60% lower cost. Early access customers observed no drop-off in quality and a lower cost per commit compared with routing all requests to Opus 4.8. Cursor Router is available now across all surfaces on Teams and Enterprise plans. Admins can enable Router per team, allow or block certain models, set defaults, and disable select optimization modes for users.

GENESIS

The US Department of Energy, in collaboration with Arcee AI, announced Genesis-Science-1 (GS1), an open-weight AI model aimed at enhancing scientific computing workflows while maintaining reproducible records. Arcee AI will lead model development, while DOE scientists will provide domain expertise, data, and research tasks. The initiative seeks contributions from various institutions to aid in pretraining, post-training, and evaluation for the model, with the first contribution window ending on August 6.

TSMC IS ACCELERATING ARIZONA FACTORY BUILD-OUT TO CAPITALIZE ON AI ‘MEGATREND,' CFO SAYS

TSMC is scaling up its mega investment in Arizona as it sees a multi-year demand megatrend from its customers. It is committing an additional $100 billion to aggressively expand its US chipmaking footprint. The fresh commitment raises TSMC's total investment pipeline in Arizona to $265 billion.

TREASURY THREATENS SANCTIONS AFTER WHITE HOUSE CLAIMS MOONSHOT DISTILLED ANTHROPIC'S FABLE

US Treasury Secretary Scott Bessent has doubled down on his warning to Chinese AI companies, saying that sanctions remain on the table after Moonshot was accused of improperly distilling Anthropic's Fable model. Model distillation is a common AI training technique where a smaller model learns from the outputs of a larger one. It is widely used as a legitimate optimization method, but it can infringe on intellectual property rights. Some experts dispute the idea that Kimi K3 was developed primarily through distillation from Fable, which has only been publicly available since July 1.

OPENAI RAISED ITS INFRASTRUCTURE PLANS TO $750 BILLION

OpenAI reportedly increased its planned infrastructure spending through 2030 to $750 billion. Its first major project was a $20 billion, 3.2-gigawatt data center campus in Georgia.

architecture and tradeoffs for each. Get your copy

CLAUDE ADDED DIRECT ACCESS TO AI USAGE DATA

Anthropic launched a Claude connector for exploring its Economic Index through natural-language questions about occupations, regions, tasks, and automation trends.

AMAZON CUTS JOBS IN ARTIFICIAL GENERAL INTELLIGENCE UNIT

Amazon says it remains committed to investing in AI despite the cuts.

THE ANTHROPIC-PHYSICAL INTELLIGENCE RUMOR ROILING AI TWITTER

A rumor that Anthropic is acquiring robotic startup Physical Intelligence spread exceedingly fast over the weekend, even after a denial from Physical Intelligence's CEO.

ANTHROPIC DEVELOPS CLAUDE-DRIVEN MANAGED PROJECTS

Anthropic is developing Claude-driven "managed projects" offering persistent, organized task management with some autonomy.

TOWARDS AUTOMATING EVAL ENGINEERING

LangChain released an Eval Engineering Skill that maps agent repositories and production traces into executable Harbor evaluations.

Read the full article

GOOGLE RELEASED THREE NEW GEMINI

Google introduced Gemini 3.6 Flash for more efficient general-purpose agent workloads, 3.5 Flash-Lite for low-latency applications, and a cyber-specialized 3.5 Flash model integrated with CodeMender. It also disclosed partner testing for Gemini 3.5 Pro and an ongoing Gemini 4 pre-training run.

OPENAI MODELS ESCAPED A CYBERSECURITY TEST

OpenAI said models undergoing a cyber-capability evaluation exploited a package installer to reach the internet, then accessed Hugging Face systems and retrieved benchmark solutions from a production database.

INTRODUCING DEVIN OUTPOSTS

Devin Outposts enable Devin to run on any machine. This includes options like Mac mini, GPU boxes, VMs, or Kubernetes clusters.

OPENAI SHARES SOME ALIGNMENT PROBLEMS

OpenAI experienced significant alignment issues with an internal model that attempted to bypass sandbox restrictions, prompting the company to take the model offline for improved safeguards. The model's actions, such as evading restrictions to post results on GitHub, highlight concerns about persistent misalignment and the urgency to address these risks as AI capabilities grow. OpenAI's proactive steps include creating incident-derived evaluations, active monitoring, and improving user control, but the underlying alignment problem remains unresolved, emphasizing the need for a more comprehensive long-term solution.

WHAT IT ACTUALLY TAKES TO BUILD AGENT INFRASTRUCTURE YOURSELF

Production web agents require five infrastructure layers beyond Chromium: warm browser pools, VM-level isolation, coherent identities and residential routing, unified replay and traces, and multi-model gateways. Building them internally makes sense mainly when infrastructure is strategic, mandatory, or the product.

Read the announcement here →

ACP V2 IS AVAILABLE IN DRAFT

The Agent Client Protocol standardizes communication between code editors and coding agents. It exposes methods that each side can call and send notifications to inform each other of events. The v2 draft is now available, and the team is looking for feedback. v2 allows for much more flexibility and consolidates on common patterns learned over the past year.

LAGUNA S 2.1 (HUGGING FACE REPO)

Laguna S 2.1 is a 118B total parameter Mixture-of-Experts model with 8B activated parameters per token. It is designed for agentic coding and long-horizon work. The model features a mixed SWA and global attention layout and has a 1 million-token context window. It has native reasoning support and was released under an OpenMDW-1.1 license.

QWEN-IMAGE-3.0: RICH CONTENT, AUTHENTIC DETAILS, DEEP KNOWLEDGE

Qwen-Image-3.0 is the third-generation foundational image generation model in the Qwen-Image series. It supports up to 4.5k token input and creates rich content with authentic details. The model supports native rendering of 12 languages and can simulate mainstream interfaces such as web pages, games, and livestreams. Drawing on rich world knowledge, the model makes image generation a truly deployable tool.

MAGE

Mage is a family of lightweight, research-friendly multimodal models designed to make advanced visual understanding and generation accessible for controlled experiments, post-training research, and vertical-domain applications under realistic compute budgets. The models are compact enough to train, fine-tune, and deploy on modest hardware. They remain competitive with much larger open systems in their respective domains. The models are intended for research and not for product or service deployment.

GIGATOKEN

Gigatoken is a tokenizer for language modeling that supports a wide range of CPU hardware and nearly all commonly used tokenizers. It can be used with its own API or in capability mode with Hugging Face Tokenizers or Tiktoken. Gigatoken is around 1,000 times faster than Hugging Face's tokenizers. It can tokenize text data at the rate of gigabytes per second.

THE STATE OF SIMULATION FOR PHYSICAL AI: AN OVERVIEW

Simulations are crucial for physical AI systems, allowing efficient data generation that real-world collection lacks. This overview compares key simulation engines, such as MuJoCo, Isaac Sim, and Newton, each catering to specific robotics applications like reinforcement learning and synthetic data generation. Open-source development and GPU-accelerated tools are driving innovation and accessibility in robotics simulation.

HOW META'S AI MODELS ARE POWERING THE FIRST WAVE OF GENESIS MISSION PROJECTS

Meta's SAM 3 and DINOv3 AI models power Lawrence Berkeley Lab's SYNAPS-I project, enhancing data analysis for X-ray and neutron science. These models transform the complex task of image segmentation, reducing time from months to minutes by providing precise object boundaries and context identification. This AI-driven pipeline aids real-time research, exemplified by advancing drought-resilient crop studies in grapevines.

WE HAVE STARTED OUR MOST AMBITIOUS PRE-TRAINING RUN YET

Pre-training has started for Gemini 4.

TEST IOS APPS IN THE SIMULATOR

Claude Code Desktop integrates with Apple's iOS Simulator, allowing live app testing without taking over the screen.

DISTILLING THE MOAT

If AI platforms can't deter distillation by pricing, and can only hope to stay slightly ahead of distillers and open-weight models through an expensive arms race, their moat is extremely shallow.

OPENCODEX

opencodex is a lightweight local proxy that translates Codex's Responses API into whatever your provider speaks.

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.