Search the archive

INSIDE ROBLOX'S BET ON WORLD MODELS

Roblox is exploring the use of video models to create photorealistic experiences without compromising scale or multiplayer performance. An early version of Roblox Reality is expected later this year or early next. This post looks at how the company is bringing this vision to life. Roblox's approach doesn't require a video model to solve everything on its own. It gives the model an engine that remembers the world, enforces its rules, and keeps it true for everyone playing.

Free basic plan for users [12]→

subscribers [12]→

SOFTWARE FACTORIES, LIGHT AND DARK

A software factory is harnessed loops at scale. Running the loop with humans (light factory) increases judgment and concentration while lowering speed and breakage. Using only AI agents (dark factory) allows agents to handle the whole process, but risks a lack of understanding of the produced software. The hardest job now is knowing which checks to build and how much autonomy to delegate.

MODELS ARE WORSE AT REVIEWING THEIR OWN CODE

Claude Code and Codex are both better at finding bugs in each other's code than their own. Models produce the same types of bugs that they are most likely to miss in review. Developers can take advantage of this finding by routing reviews to different models.

OPENAI'S AGENTS REACH 10 MILLION USERS AFTER CHATGPT WORK DEBUT

OpenAI now has 10 million people using Codex and ChatGPT Work. The company has seen a sharp uptick in demand for its AI agents since the release of ChatGPT Work. It has nearly doubled agent usage from earlier this month. OpenAI is pushing to develop advanced AI agents that can streamline more complex tasks.

GOOGLE EXPANDS GEMINI LINEUP WITH CHEAPER MODELS AND NEW MYTHOS RIVAL

Google has released Gemini 3.5 Flash Cyber, Gemini 3.6 Flash, and Gemini 3.5 Flash-Lite. Gemini 3.5 Flash Cyber is designed to detect and patch software vulnerabilities and will initially be available only to governments and trusted partners through a limited-access pilot. Gemini 3.6 Flash has improved coding, multimodal, and knowledge-work performance while using up to 17% fewer tokens and costing less per token than the previous model. Gemini 3.5 Flash-Lite is built for high-volume workloads and smaller tasks within larger AI-agent systems.

CLAUDE CODE CAN NOW BUILD AND TEST IOS APPS IN APPLE'S SIMULATOR

Claude Code for desktop has been updated to work with the iOS Simulator.

A FIRESIDE CHAT WITH CAT AND THARIQ FROM THE CLAUDE CODE TEAM

Cat Wu and Thariq Shihipar from Anthropic's Claude Code team talk about Claude Code, Claude Tag, Fable, coding agent security, evals, tool design, and how Anthropic uses these tools.

A NEW SOFTWARE ENGINEERING PARADIGM

The combination of formal verification and AI is much more effective than AI by itself.

THE GEOPOLITICS OF OPEN WEIGHTS

While China has a reputation for releasing open-weight models, not all Chinese companies are actually following the same approach.

CLAUDE IS NOT A COMPILER

Claude is better than a compiler because it can work vertically across the stack.

LLMS ARE SURPRISINGLY BAD BLOG AUTHORS

Nobody likes reading AI slop.

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.

GOOGLE IS OPEN-SOURCING ITS 3D EMOJI

Google is releasing its 3D emoji as open-source .OBJ files, letting anyone use them for VR worlds, apps, or memes. The announcement came alongside design details shared for World Emoji Day, addressing dimensional questions like whether a smiley is a sphere, mask, or disc. The 3D emoji set, called Noto Emoji 3D, debuted in May to mixed reactions.

ADOBE'S PROJECT INDIGO CAMERA APP CAN NOW EDIT AND CRITIQUE YOUR PHOTOS WITH AI

Adobe's Project Indigo camera app is testing AI Playground, adding AI editing tools like distraction removal, restyling, and a photo critique feature. The tools run on Google's Nano Banana model, capping edited images at 2K pixels, with Adobe planning to add its own Firefly models later. Access is free but limited to a small percentage of users in version 1.1 for a few weeks, with wider availability still undecided.

APPLE'S NEXT BIG IPHONE DISPLAY EXPERIMENT COULD BE ITS BIGGEST YET

Apple is reportedly testing a larger 6.96-inch display for a future iPhone, bringing it close to the 7-inch mark while maintaining the same aspect ratio as the current Pro Max. The screen could be part of Apple's ambitious 20th-anniversary plans for 2027, but there's no evidence yet that it will appear in a final product. As with all early hardware leaks, the display remains experimental and could be changed or scrapped before release.

FROM VIBE TO SPECS: RECLAIMING THE DESIGN PROCESS WITH SAID FRAMEWORK

AI-native product teams are replacing linear design handoffs with the "SAID framework," a continuous loop built around specs, Git, and human-agent collaboration. Rather than fixed roles, teams now shift between archetypes—Prototyper, Builder, Sweeper, Grower, and Maintainer—while specs, tested via Anthropic's 4D Fluency Framework (Delegation, Description, Discernment, and Diligence), become the shared source of truth. Designers aren't expected to become engineers, but must learn to express intent as precise, testable behavior that both humans and AI agents can execute.

TRAINING DESIGN SENSES

Design judgment and taste must be deliberately trained through practice, not measured by years of passive experience in the field. This post outlines four trainable areas—observation, hand-eye dexterity, aesthetic fluency, and rhetoric—that build intuition through repeated, intentional effort. Genuine design intuition emerges from accumulated evidence of practice, not from time alone, and must be demonstrated through actual work.

MAKE IT WORK VS. MAKE IT GOOD

Making something work is often easier to recognize and reward than making it truly good, even though the latter requires far more skill and effort. Great design comes from refining usability, accessibility, performance, and overall experience until the solution feels obvious rather than merely functional. As the saying goes, "make it work" is the first 90% of the job, while "make it good" is the other 90%.

MAKE EVERY VIDEO WORTH WATCHING

Dina records your screen and refines it automatically — so what your audience sees is as clear as what you imagined.

SYMBOLS THAT SPEAK COLOR

Colorsym is a free, open-source way to identify color when color alone isn't enough. Memorable symbols for base and combined colors in games, print, signage, and more.

AI MOTION DESIGNER AND AI ANIMATOR PLATFORM

Create professional motion graphics, animated typefaces, and stunning video animations without After Effects. Export to MP4, MOV, and GIF instantly.

NOT ALL BRAND REDESIGNS ARE TRAINWRECKS

While Cracker Barrel, Jaguar, and Tropicana suffered backlash for their redesigns, several other brands managed rebrands that pleased audiences and critics alike. Examples include the Premier League's simplified lion logo, Dunkin's name change dropping "Donuts," Rolls-Royce's sleeker Spirit of Ecstasy emblem, Reddit's unified pre-IPO redesign by Pentagram, and eBay's decluttered 2024 website. These successful makeovers contrast with more infamous rebrand failures, showing that thoughtful redesigns can modernize a brand's image without alienating its existing customer base.

THE SWISS CHEESE MODEL, PT. 1

The Swiss Cheese model explains how major failures often result from several small, independent mistakes aligning rather than a single catastrophic error. A famous example is Knight Capital's $460 million trading loss, where outdated code, a reused software flag, an incomplete deployment, missing verification, and inadequate safeguards combined to trigger a company-ending failure. The key lesson is that robust systems rely on multiple overlapping protections, because any one small mistake is usually harmless until several failures line up at once.

HOLOGRAPHIK'S IDENTITY FOR VEDA TREATS COINS AS ANCIENT ARTEFACTS

Instead of relying on the generic visual language common in crypto, Holographik created a distinctive identity for Veda by reimagining the familiar coin as a symbolic artifact inspired by ancient stores of value and knowledge. Each 3D coin represents the underlying architecture of a specific crypto vault, combining custom illustrations, metallic materials, and restrained typography to balance trust, memorability, and sophistication. The result is a brand system that feels tangible and timeless while standing apart from typical fintech aesthetics.

WHY SO MANY AI COMPANIES' LOGOS LOOK LIKE…THAT

COCA-COLA GETS NEW LOOK THAT MAKES IT EVEN "MORE COCA-COLA"

Coca-Cola unveiled a global rebrand by agency Jones Knowles Ritchie, aiming to make the brand "more Coca-Cola" across over 200 markets.