Search the archive

Subscribe now.

A WORD ABOUT COMPLEXITY

One of the barriers that prevents people from modifying open source software is complexity.

BECOMING THE PERSON WHO DOES THE THING

Our self-identity dictates everything, but it isn't set in stone.

WHY YOU'D ISSUE A BRANDED STABLECOIN LIKE MCDONALDSCOIN

Branded stablecoins entitle platforms to more of the issuer rewards for customer deposits, which consequently lets them pay more rewards to customers who choose to store money with them.

XAI LAYS OFF 500 DATA ANNOTATORS

xAI has reportedly laid off a third of its data annotation team as it pivots to expand its specialized AI tutor division.

XAI LAUNCHES GROK 4 FAST IN EARLY ACCESS BETA WITH UP TO 10X SPEED

Grok 4 Fast, the newest addition to xAI's lineup, is now available for users on the Grok web interface via the model selector. It can be accessed by enabling a new toggle in the Subscription settings. Marked as an early access beta, Grok 4 Fast is up to 10 times quicker than the standard Grok 4. It is optimized to respond rapidly by spending minimal processing time on complex tasks, which limits its creative abilities.

POST-TRAINING 101 FOR LLMS

A walkthrough of the entire post-training lifecycle of LLMs, from supervised fine-tuning and reward modeling to reinforcement learning methods such as RLHF, along with evaluation best practices.

THE VERTICAL AI PLAYBOOK (BOOK)

Despite billions invested, 42% of enterprise AI initiatives were discontinued in 2024. This was caused by how models were embedded into business. The winners redesign workflows, rethink structures, and take ownership of the service layer where value is created. The next generation of CEOs will treat AI as a labor class and deploy the technology with the same discipline that the most successful serial acquirers apply to capital.

BREAKING GPT-OSS: A BRIEF INVESTIGATION

This article evaluates different jailbreaking methods against gpt-oss. The model appears to have had robust safety training in both system prompting and refusal vector attacks. It is tricky to work with, and not all libraries support its idiosyncrasies.

workloads. Find out why Google was named a leader

VAULTGEMMA: THE WORLD'S MOST CAPABLE DIFFERENTIALLY PRIVATE LLM

VaultGemma is a model that Google trained from scratch with Differential Privacy (DP). DP offers a mathematically robust solution to user privacy that adds calibrated noise to present memorization. It has some trade-offs, like reducing training stability and significantly increasing batch size. There is still a utility gap between DP-trained and non-DP-trained models, but that gap can be systematically narrowed with more research on mechanism design for DP training.

THE ILLUSION OF DIMINISHING RETURNS: MEASURING LONG HORIZON EXECUTION IN LLMS

Real-world value often stems from the length of a task an agent can complete. Marginal gains in single-step accuracy can compound into exponential improvements in the length of a task a model can successfully complete. Models are more likely to make mistakes when the context contains errors from previous turns. Failures when tasks are made longer arise from mistakes in execution rather than an inability to reason.

THE SECOND WAVE OF MCP: BUILDING FOR LLMS, NOT DEVELOPERS

Teams that shift from API shaped tools to workflow-shaped tools see meaningful improvements in reliability and efficiency. MCP works best when tools handle complete user intentions rather than exposing individual API operations. Large language models don't work like developers - they have to constantly rediscover which tools exist, how to use them, and in what order, so building tools around workflows produces better results.

YOU SHOULD BE REWRITING YOUR PROMPTS

Models aren't perfectly interchangeable - if you are switching models, rewrite your prompts. Prompts overfit to models the same way models overfit to data. They need to be tested, evaluated, and aligned with the defaults of the new model. Adapting prompts will save tokens while producing better results.

AI WILL NOT MAKE YOU RICH

Most of the new value created by AI will be captured by consumers, who will see wider and more affordable access to services like medical care, education, and advice. Knowledge-intensive services will get cheaper, allowing consumers to buy more of them. At the same time, services that require person-to-person interaction will get more expensive and take up a greater percentage of household spending. There will be obvious opportunities in both. Think through the implications of knowledge workers becoming more efficient, imagine what markets this efficiency unlocks, and invest in those.

perfect AI coding environment. Try Warp Code for free

UNDERSTANDING GPU ARCHITECTURE

Cornell's Center for Advanced Computing published an interactive workshop covering GPU memory hierarchies, streaming multiprocessors, and detailed breakdowns of NVIDIA's Tesla V100 and Quadro RTX 5000 architectures.

MANAGING AGENT MEMORY WITH SESSIONS

How to manage short-term memory for AI agents using the OpenAI Agents SDK, employing trimming and compression techniques to keep sessions coherent, fast, and reliable.

NVIDIA STEPS BACK FROM DGX CLOUD — STOPS TRYING TO COMPETE WITH AWS AND AZURE

Nvidia now uses its DGX Cloud capacity for internal research.

OPENAI GROVE PROGRAM ANNOUNCEMENT

OpenAI has announced a 5-week residency for early-stage technical founders, offering mentorship, early tool access, and peer collaboration to explore new AI product ideas.

POLYMARKET AND KALSHI EYE MULTI-BILLION VALUATIONS

Prediction market giants Polymarket and Kalshi are preparing new funding rounds at high valuations. Polymarket is planning to raise at $9-10B, and Kalshi at $5B. Polymarket, fresh off receiving CFTC clearance to relaunch in the U.S., is scaling quickly with global pseudonymous trading on Polygon and USDC settlements, while Kalshi remains a CFTC-regulated U.S. platform requiring fiat deposits and KYC.

NATIVE MARKETS CLAIMS HYPERLIQUID'S USDH TICKER

After a contentious community voting process, Native Markets has secured the USDH ticker for Hyperliquid's US dollar-pegged stablecoin. The rollout will begin with a testing phase featuring capped transactions, followed by the opening of the USDH/USDC spot order book and the removal of transaction caps. The selection process has drawn criticism from industry executives who alleged a predetermined outcome favoring Native Markets, suggesting controversy around what was presented as an open competitive process among multiple stablecoin issuers, including established players like Paxos, BitGo, and Ethena.

TETHER ANNOUNCES PLANS TO LAUNCH USAT, A U.S.-COMPLIANT STABLECOIN

Tether will launch USAT, a U.S.-compliant stablecoin issued through Anchorage Digital and backed by Cantor Fitzgerald, marking its first regulated entry into the American market. Designed as a “hedge-and-expand” strategy, USAT provides a compliance-first option for banks, fintechs, and corporates under the GENIUS Act, while USDT continues to serve offshore, permissionless markets. The move directly challenges Circle's USDC by eroding its positioning as the regulated dollar token of choice, raising the stakes in the ongoing battle for distribution and integrations.

ETHEREUM FOUNDATION SETS END-TO-END PRIVACY ROADMAP

The Ethereum Foundation's Privacy & Scaling Explorations team has rebranded to Privacy Stewards of Ethereum (PSE), releasing a roadmap for building full end-to-end privacy on Ethereum. The plan focuses on private writes (cheap, seamless private transactions), private reads (blockchain access without exposing user identity or intent), and private proving (fast, accessible zero-knowledge proofs).

A TECHNICAL TYPOLOGY OF PREDICTION MARKETS

Prediction markets break down into five key design dimensions, revealing the complex tradeoffs platforms make when balancing decentralization, efficiency, and user experience. This typology covers outcome resolution methods (oracle-based, like Polymarket vs human-reviewed, like Kalshi vs automated feeds), liquidity models (AMMs used by XO Market vs order books used by evolved Polymarket vs hybrid approaches by Kalshi), event creation mechanisms (platform-curated vs user-generated vs community-vetted), market types (binary vs multi-outcome vs scalar markets like Trepa's precision-based approach), and underlying infrastructure (off-chain vs onchain vs hybrid). Analysis reveals that successful prediction markets must solve the fundamental liquidity problem while making strategic choices about centralization versus decentralization, with platforms like Kalshi prioritizing regulatory compliance and institutional liquidity, while fully onchain protocols like Augur emphasize censorship resistance and trustlessness. There is no single "best" prediction market design, as the optimal approach depends entirely on target audience, use case, and tolerance for trade-offs, with innovation continuing across multiple dimensions as the ecosystem matures.

FED, DATS, AND NPM

Coinbase Institutional's Weekly Market Commentary anticipates a 25 basis point Federal Reserve rate cut on September 17, driven by weakening labor and housing markets, though persistent core services and shelter costs make a more aggressive 50 basis point cut unlikely despite August's 0.4% month-over-month CPI increase. Digital Asset Tokens (DATs) have entered a "Player versus Player" phase with market Net Asset Values compressing toward 1 and trading volumes cooling, though NAV and supply share continue rising, indicating sustained institutional interest despite reduced speculative premiums. There is intense competition for Hyperliquid's USDH stablecoin partnership, with issuers proposing to share 95-100% of reserve yields, potentially redirecting $200-220 million annually and enhancing HYPE token value through buybacks and burns, with a final vote scheduled for September 14.

FIXED YIELD WILL BE DEFI'S TROJAN HORSE FOR TRADFI

Fixed yield products represent DeFi's institutional adoption strategy by offering predictable cash flows that traditional finance already understands, with Ethereum-based fixed yields delivering 8-15% APY compared to T-Bills' 3-5% returns across a $33.2 trillion market. Pendle's yield trading mechanism splits yield-bearing assets into yield tokens (YT) and principal tokens (PT), mirroring the familiar bond structure of separating coupons from principal while leveraging code enforcement and onchain settlement rather than relying on distressed issuers or opaque IOUs. The yield trading market has scaled to over $10 billion in TVL with $100 million in daily volume, with Pendle settling more than $54 billion in fixed yield derivatives across ETH derivatives, stablecoins, and real-world assets. Institutional demand for stable, predictable onchain yield is expected to grow following the GENIUS Act passage and corporate exploration of stablecoins by companies like Walmart and Amazon, positioning fixed yield as the anchor for onchain finance, similar to how Treasuries anchor traditional finance.

STABLECOINS FLIP THE REMITTANCE MODEL

Stablecoins are transforming remittances by letting senders transfer USD directly, leaving recipients to decide when (or if) to convert into local currency. This shift offers greater flexibility, stability, and utility while moving FX revenues downstream from traditional remittance firms to wallets and neobanks. In markets like Argentina, businesses already hold stablecoin treasuries, signaling where consumer adoption is headed.

ASTROLOGY: AI-DRIVEN HOROSCOPES ON BASE

Astrology, a new mini app on Base, lets users mint zodiac-themed art and receive personalized AI-powered daily horoscopes.

PATH TO ALTSEASON

In tracking crypto market cycles, there are four phases of capital rotation.

TRANSACTION FEES NO LONGER A BARRIER TO BLOCKCHAIN ADOPTION

Average transaction fees on major chains like Ethereum (~$0.50), Bitcoin (~$1.40), and Solana (near $0) have dropped to levels where they no longer block mainstream use.

SOLANA STABLECOIN ALL-TIME HIGH

Solana's stablecoin TVL has reached a new all-time high of $13.3 billion, according to data from Artemis.

FIGMA LATEST UPDATES

Figma updates include instant frame transfers from Design to Buzz, a restored Community tab for browsing public resources, and a new Resources tab for internal team templates. Figma Make now supports editable templates. Apple UI kits for watchOS, macOS, and visionOS have been added but are disabled by default.

ADOBE'S STOCK GAINS AS ITS AI INVESTMENTS START BEARING FRUIT

Adobe's stock rose in after-hours trading following third-quarter results that exceeded Wall Street expectations, with earnings of $5.31 per share and revenue of $5.99 billion. The company's AI investments are generating significant returns, with annual recurring revenue from AI products reaching over $5 billion and "AI-first" ARR surpassing the $250 million target. Adobe raised its full-year revenue forecast to $23.65-23.7 billion, demonstrating strong momentum despite facing competition from AI-native startups like Canva and Figma.

ROBLOX ANNOUNCES NEW AI TOOLS AND SHORT-FORM VIDEO PLATFORM FOR CREATORS

Roblox unveiled Roblox Moments, a TikTok-like short-form video platform launching in beta for users 13+ to share and watch 30-second gameplay clips. The company also introduced new AI tools enabling creators to generate interactive vehicles and weapons through text prompts, plus real-time voice chat translation across four languages. Additional AI features include text-to-speech APIs with voice presets for character dialogue and speech-to-text capabilities for vocal game controls.

DESIGNING FOR AI AGENTS AND UX PATTERNS

AI agents are autonomous systems that perceive environments, make decisions, and execute multi-step workflows independently, unlike traditional AI tools that simply respond to prompts. This article explores key UX design challenges, including trust, control balance, and contextual awareness, while presenting four core patterns: Progress Reveal (showing agent reasoning), Context Window (displaying information sources), Teach-Back (explaining learning from corrections), and Collaborative Workspace (enabling real-time human-AI collaboration). Successful agent experiences focus on building relationships through personality consistency, learning loops, boundary setting, and recovery processes that strengthen user trust.

MOTION DESIGN ISN'T JUST EYE CANDY; IT'S UX INFRASTRUCTURE

Motion design has evolved beyond visual appeal to become an essential UX infrastructure that builds trust, conveys messages quickly, and enhances user comprehension through strategic animations. Modern tools like LottieFiles and the dotLottie format enable teams to implement lightweight, scalable animations without compromising performance, as demonstrated by companies like CNN Create, Gojek, and Wise. While AI can support the creative process with layout suggestions and design recommendations, effective motion design still requires human empathy and brand understanding to create emotionally resonant user experiences.

DESIGNERS, WHY DO WE ACCEPT BEING TREATED AS CHILDREN?

Modern software designers often end up translating business needs into visuals without real influence, leaving developers to shape the product. To regain authority, designers need to move beyond static mockups and engage directly with building, testing, and shipping real products, closing the feedback loop and practicing true design.

BRAND PERFORMANCE ON AI SEARCH PLATFORMS

Track, analyze, and improve brand performance on AI search platforms through key metrics like visibility, position, and sentiment.

A FAST AND MODERN SVG EDITOR

Hyvector is an SVG editor that runs in all modern browsers. It's stable, very fast, and capable of handling complex SVG images.

THIS RESTAURANT'S BRANDING IS LIKE A PORTAL TO MEXICO

Mixteca, a New York Mexican bar and restaurant, has unveiled a new brand identity by Grandmerci that blends authentic Mexican roots with modern design. Featuring custom typography, Talavera-inspired patterns, and rustic textures, the identity captures the warmth and spirit of a Mexican town while creating a vibrant, contemporary dining experience.

THE ARCHITECTURE OF HUMAN ERROR

Good design isn't just about functionality or user experience, it must also reduce human error and support recovery when mistakes occur. The SRK model categorizes errors as skill-based (slips and lapses), rule-based (misapplied logic), knowledge-based (judgment under uncertainty), or violations (intentional deviations). Each type requires different prevention strategies, highlighting that human error is a product of design and should be addressed through systems that align with cognitive processes.

HOW NOUVELLE IDÉE DESIGNED THEIR OWN SANDWICH SHOP IN MONTREAL

A new Montreal sandwich shop identity balances personal investment with collaborative creativity, resulting in a brand that feels authentically local and timeless. With a hand-drawn wordmark, utilitarian supporting typography, a sports-inspired colour palette, and design cues from vintage diners and athletics, the space is crafted to feel like a long-standing neighbourhood institution.

NVIDIA RECKONS ITS NEW AI PIPELINE IS THE END OF TEDIOUS PROTOTYPING FOR 3D ARTISTS

Nvidia launched its AI Blueprint for 3D Object Generation at IFA Berlin, allowing 3D artists to populate scenes using simple text prompts instead of manual prototyping.

OPENAI BACKS AI-GENERATED ANIMATED FILM ‘CRITTERZ'

OpenAI is backing "Critterz," an AI-generated animated film created with its tools, including DALL-E and Sora, aiming to complete the feature in nine months for under $30 million.

SARA MAESE IS CREATING AN ENTIRE UNIVERSE OUT OF SIMPLE FORMS AND LOVELY WARM COLOURS

Spanish illustrator Sara Maese creates playful, pastel-toned artworks where simple shapes and bold lines form lively characters and whimsical scenes.