Search the archive

ALL YOUR COWORKERS ARE PROBABILISTIC TOO

People get things wrong, just like large language models.

'BABY SHARK' CREATOR PINKFONG SHARES SURGE UP TO 62% ON DEBUT

Pinkfong went public to expand into a full-fledged media studio.

MIDDLEMEN ARE EATING THE WORLD (AND THAT'S GOOD, ACTUALLY)

Middlemen help people connect to work on endeavors that would not otherwise be possible.

HOW YOUR BRAIN CREATES ‘AHA' MOMENTS AND WHY THEY STICK

Insight involves abrupt cognitive shifts in how the mind understands information.

PRE-PEP: RUST FOR CPYTHON

Rust provides many extremely useful benefits that improve CPython development, like increasing memory safety.

Start with 100 free credits

GROK 4.1

Grok 4.1 is the top model on LMArena. xAI touted its large-scale post-training infrastructure as responsible for targeted improvements to emotional intelligence and creative writing, which suggests it is pursuing consumer-focused AI companion use cases currently dominated by ChatGPT.

ARM CUSTOM CHIPS GET A BOOST WITH NVIDIA PARTNERSHIP

Arm-based Neoverse CPUs can now be paired with Nvidia's GPUs using Nvidia's NVLink Fusion technology. Microsoft, Amazon, and Google are all developing or deploying Arm-based CPUs in their clouds. The partnership will make it easier for customers of both companies to pair hardware together. The announcement signals that Nvidia has decided to open up its NVLink platform instead of forcing customers to use its CPUs.

CUDA, SHMUDA: RUNNING ALPHAFOLD ON A MACBOOK AIR

Apple's MLX framework is great for science workloads because it was designed from the ground up to take advantage of Apple's unified memory architecture. Models like AlphaFold are memory-intensive, and almost all protein software is heavily optimized for CUDA. This post shows how a developer got AlphaFold to run on an Apple M chip machine. ARM chips like the Apple M are extremely fast and use only a fraction of the wattage that a single CUDA card needs, and getting a Mac device is a lot easier than getting a Blackwell GPU.

ROADMAP FOR VIDEO-BASED WORLD MODELS

This roadmap outlines the progression of video generation systems into full-fledged simulators that combine internal physics-aware world models with visual rendering. It defines four developmental stages toward interactive and stochastic virtual environments.

RL IS EVEN MORE INFORMATION INEFFICIENT THAN YOU THOUGHT

It takes way more FLOPs to get a single sample in RL than it does in supervised learning. RL requires unrolling a whole thinking trajectory tens of thousands of tokens long to get a single reward signal at the end. In pretraining, you get a signal on every single token you train on. For most of training, the information density per sample is way lower for RL compared to supervised learning.

THE AGENT LABS THESIS

Agent Labs primarily research and sell agents. They put product first and use outcome-based pricing, as opposed to Model Labs, which put models first and price per token. Agent Labs have better cashflow economics, but it might take longer to see exit valuations. The Model Lab mission may be shifting, at least until the next big algorithm shift.

WEATHERNEXT 2: OUR MOST ADVANCED WEATHER FORECASTING MODEL

WeatherNext 2 is an AI model that delivers efficient, accurate, and high-resolution global weather predictions. It can generate forecasts with a resolution of up to an hour. The model's forecast data has been released. WeatherNext 2 will be integrated to help power weather information in Google Maps in the coming weeks.

BUILD A GPT-5.1 CODING AGENT

This guide walks through building a coding agent using GPT-5.1 and the Agents SDK, leveraging tools like shell execution, patch editing, web search, and Context7 MCP for live documentation access.

INTELLIGENCE PER WATT: MEASURING INTELLIGENCE EFFICIENCY OF LOCAL AI

Frontier models today are served through massive, centralized cloud computing services that are straining electric grids. The performance gap between these large models and much smaller models, the kind that can run on a phone or laptop, is shrinking.

AA-OMNISCIENCE

AA-Omniscience is a benchmark for knowledge and hallucination across more than 40 topics. It punishes hallucinations by deducting points when models guess rather than admitting they do not know the answer. The benchmark complements the Artificial Analysis Intelligence Index by incorporating the measurement of knowledge and probability of hallucination. This thread contains the benchmark results from testing cutting-edge models against AA-Omniscience.

REPLICATE IS JOINING CLOUDFLARE

Replicate, a platform for running AI models, is joining Cloudflare. Existing Replicate users will be able to continue work without interruption. Existing Workers AI users will soon receive a massive expansion of the model catalog and the new ability to run fine-tunes and custom models directly. Cloudflare's AI Gateway will provide users with a single control plane for observability, prompt management, A/B testing, and cost analytics across all of their models, whether they're running on Cloudflare, Replicate, or any other provider.

November 20

GOOGLE TO ENABLE RESEARCH AUTOMATION ON GEMINI ENTERPRISE

The unreleased system generates roughly 100 ideas on a topic, then spawns agent teams that compete in a tournament-style bracket for the best result.

OPENAI IS FINALLY LETTING EMPLOYEES DONATE THEIR EQUITY TO CHARITY

Current and former OpenAI employees with eligible shares will be able to donate their shares for a short period.

DATABRICKS SEEKS $130 BILLION VALUATION

Databricks is in talks to raise funds at a valuation above $130 billion, though nothing has been signed yet.

NEW WAYS TO PLAN TRAVEL WITH AI IN SEARCH

Canvas in AI Mode now builds complete travel itineraries by pulling together flight data, hotel comparisons, Maps reviews, and restaurant recommendations.

CRYPTO FUNDS SEE $2B WEEKLY OUTFLOWS

CoinShares reported $2 billion in outflows from global crypto investment products, the largest weekly exit since February, as shifting US rate cut expectations and selling from large holders dampened sentiment. The US led with 97% of redemptions, while Germany bucked the trend with $13 million in inflows. Bitcoin funds saw $1.38 billion in withdrawals and Ethereum $689 million, while multi-asset ETPs attracted $69 million as investors sought diversified exposure amid macro-driven volatility.

BITCOIN'S 25% SLIDE REFLECTS SHORT-TERM CORRECTION

Bitcoin's 25% drop from its October 6 all-time high of $126,000 reflects investor anxiety over historical four-year cycle peaks rather than weakening fundamentals. 340,000 BTC sold by long-term holders was absorbed by $34 billion in ETF inflows and corporate treasuries. Institutional ETF ownership climbed from 20% to 28% with $125 billion in assets under management despite recent outflows, while Strategy's conservative leverage ratio of $8 billion debt against $61 billion in bitcoin holdings dismisses liquidation concerns. A shallow correction is expected rather than the typical 60-70% drawdown as structural tailwinds remain intact.

1INCH LAUNCHES ‘AQUA' TO UNIFY DEFI LIQUIDITY

1inch Aqua is a shared liquidity model designed to eliminate DeFi's capital fragmentation by turning every wallet into a self-custodial AMM, allowing the same capital to power multiple strategies simultaneously. Developers can now access Aqua's SDK, libraries, and bounties up to $100,000. Aqua is positioned as a foundational liquidity primitive that removes deposit and withdrawal logic while boosting efficiency across chains. The launch at Devconnect builds on 1inch's aggregation success and aims to create a capital-efficient base layer for DeFi composability.

AAVE APP: A SMARTER WAY TO SAVE

Aave is launching an app that allows users to earn up to 9% on stablecoin balances. It uses embedded wallets to create an email-based sign-up solution, mitigating the need to interact with a complex browser wallet and web interface. Aave will offer insurance on deposits up to $1 million per user. The waitlist is currently live. There is no clear launch date yet.

HOW YIELD IS ACTUALLY MADE

DeFi vaults automate yield generation through smart contracts that deposit assets, allocate to protocols like Aave and Curve, compound returns, and continuously rebalance positions across nine core strategies, including yield aggregation, delta-neutral hedging, leveraged farming, and options-based derivatives. The Stream Finance case exposed systemic risk when curator failures triggered $93 million in cascading losses across interconnected protocols, demonstrating how recursive vault loops and poor management amplifies contagion beyond protocol boundaries. Vaults simplified DeFi from complex multi-step strategies into single-action deposits, but the next phase requires stricter audits, DAO-based curator approvals, and transparent proof-of-reserves dashboards to balance automation with accountability as smart contract bugs and oracle manipulation remain persistent attack vectors.

WHY YOUR COIN ISN'T PUMPING

Crypto prices remain depressed despite ETF launches and institutional adoption because blockchains trade at software multiples on casino revenue, with Ethereum at 200-400x price-to-sales on $1-2 billion in cyclical fees versus Nvidia's 40-45x price-to-earnings with recurring contracts. The industry deployed over $100 billion into infrastructure, while the entire crypto ecosystem serves roughly 40 million active users, compared to OpenAI's rumored $1 trillion IPO valuation serving 20x more users. Ethereum faces structural value leakage as L2s capture fees while the base layer carries security costs, creating a federal government collecting only state-level tax revenues, with most blockchain revenue tied to perpetuals, memecoins, and MEV spikes rather than stable enterprise cash flows.

ETHEREUM'S ECONOMIC AND PRIVACY EDGE OVER BITCOIN

Ethereum's rollup-centric architecture offers privacy without sacrificing composability, unlike Bitcoin or Zcash, allowing private interaction with DeFi and financial apps. ETH is also a superior store of value due to its native staking yield, which stays above net issuance, eliminating miner-driven sell pressure. Ethereum-based Digital Asset Treasuries (DATs) are therefore structured around ETH as both a yield-generating and privacy-preserving asset.

REVISITING THE HYPEREVM THESIS

Three months after predicting HyperEVM would become crypto's most aligned L1, the thesis faces market pressure as TVL dropped $500 million and HYPE fell 13.5% to $38.50 amid industry-wide airdrop farmer exhaustion following October's Black Friday crash. Native Markets launched USDH, the first stablecoin using 50% of yield for HYPE buybacks, while HIP-3 projects like Kinetiq and Ventuals deployed leveraged equities and pre-IPO markets but struggle with liquidity despite Robinhood listings and expanding builder code distribution. The approaching one-year HYPE TGE anniversary brings team unlock concerns, yet HyperEVM remains crypto's only chain where L1 token, infrastructure, and builders are economically aligned through mechanisms impossible elsewhere, like delta-neutral LP strategies combining AMMs with native perp DEX primitives.

ARTEMIS RELEASES 2025 STABLECOIN ECOSYSTEM MAP

Artemis's 2025 Stablecoin Map visualizes over 250 projects across the stablecoin landscape from issuers like Circle and PayPal to infrastructure players like Iron and BlockRadar.

EIP-7918 TO FIX ETHEREUM'S BLOB PRICING FLAW

Ethereum's current blob pricing misreads reduced blob usage during periods of high L1 gas as low demand, making blobs artificially cheap and causing L2s to halt postings.

POLYGON LEADS P2P PAYMENTS WITH 40% MARKET SHARE

THE COMPETITIVE MOAT OF ONCHAIN FIRMS

In the onchain economy, firms extracting value for insiders will lose to hyper-efficient onchain organizations that circulate surplus directly to users.

SMITH & DICTION REBRANDS SUPERHUMAN, FKA GRAMMARLY

Grammarly's rebrand into Superhuman grew from a simple AI-agent brief into a full transformation after major acquisitions and the launch of its AI assistant. Smith & Diction created Hero, a simple character-based logo whose expressive motion helps convey intelligence and adaptability across the product ecosystem. This flexible identity, supported by customised typography and deep product integration, unifies the previously separate brands and strengthens Superhuman's shift from a writing tool to a broad AI productivity platform, helping align teams and attract renewed interest.

SERVICENOW AND FIGMA LAUNCH STRATEGIC COLLABORATION TO TURN DESIGN VISION INTO ENTERPRISE TRANSFORMATION

ServiceNow and Figma have launched a strategic collaboration utilizing MCP technology, enabling developers to transform Figma designs directly into enterprise-ready applications through ServiceNow's Build Agent. The integration automates the path from design to deployment with enterprise-grade security, reducing initial UI and data model implementation time by over 80%. Users can now provide a Figma design link as a prompt to the Build Agent, which interprets layouts and components to automatically generate fully functional, secure, and scalable applications within minutes.

OPENAI IS LAUNCHING GROUP CHATS IN CHATGPT

OpenAI has introduced group chats for ChatGPT, allowing up to 20 people to collaborate with the AI in a shared conversation. Initially rolling out in Japan, New Zealand, South Korea, and Taiwan, the feature supports planning and decision-making while keeping group chats separate from private memory, adds new social behaviors like emoji reactions, and includes participant and notification controls. Broader expansion is planned for after the pilot.

FIGMA JUST PROVED EVERYONE (INCLUDING ME) WRONG

Figma's 2025 Schema Conference introduced nine major updates, including extended collections for multi-brand systems, slots for component customization, and AI-powered design checking, addressing long-standing design system needs. The announcements mark a significant shift from previous concerns about Figma neglecting its design system features, with improvements spanning performance, code integration through MCP servers and GitHub connections, and native variable import/export. Variable composability still requires further development for advanced authoring patterns.

HOW DO YOU DEAL WITH AN EXECUTIVE DEMANDING FEATURES? CREATE A HYPOTHESIS

When executives demand features based on trends, designers should respond with strategic curiosity rather than compliance or refusal, using a stakeholder-as-user approach to understand motivations and context. Designers can justify their questions by estimating the likelihood that design decisions will impact user behavior, much like engineers provide development time estimates. Summarizing ideas into testable hypotheses—"If we [change feature], users will [behavior] because [rationale], achieving [goal]"—helps expose logical inconsistencies and may lead executives to reconsider poor decisions themselves.

AFFINITY BY CANVA REVIEW: FREE IS THE MAGIC NUMBER

Affinity by Canva is now a single, completely free app that merges Designer, Photo, and Publisher into one interface with Vector, Pixel, and Layout modes, plus optional Canva AI tools for paid users. Setup is simple and the software retains its strong tutorial support, robust vector and photo-editing features, non-destructive workflows, and comprehensive PDF export, while adding downloadable on-device ML models for tasks like object selection. Despite its refreshed packaging, it's largely the same Affinity suite users know, with no major losses, though it still lacks photo cataloguing tools and older Affinity files convert to a new .af format that legacy apps can't open.

CREATE PROFESSIONAL AI VIDEOS IN MINUTES

Transform your ideas into high-quality videos in 60 seconds. Perfect for social media, marketing, and creative content.

ALL-IN-ONE AI WORKSPACE FOR CREATIVE TEAMS

Generate, edit, and enhance images, videos, and text in one intelligent whiteboard.

BLOOD SUGAR BATTLER: BUILDING A REAL GAME AS A DESIGNER USING AI

Alex Pierce built Blood Sugar Battler, a mobile diabetes education game, over the course of three months, utilizing the AI tool Lovable.dev for coding while hand-crafting all visual and audio elements. The retro-styled game challenges players to tap falling foods within 60 seconds, managing both score and a blood sugar meter that reacts differently to healthy versus processed foods. Pierce handled all creative decisions—pixel art sprites, vintage macOS interface design, sound effects, and gameplay balancing—while using AI to translate design intent into functional code. They learned that AI excels at technical scaffolding but cannot replace artistic direction or critical thinking.

SOL DUSTS OFF THE ARCHIVES IN SUNNY REBRAND

Sol has launched a refreshed identity aimed at young adults, blending its Mexican heritage with a modern, energetic aesthetic. Created by LOVE with illustrator Tobias Hall, the rebrand introduces a revitalised wordmark, custom typography, updated historical elements, and a brighter palette featuring elevated reds, golds, and new teal accents. Together with an uplifted sun icon and more confident attitude, the redesign delivers a “newstalgic” brand experience that feels both authentic and contemporary.

WHAT HAPPENS WHEN A UX RESEARCHER STARTS USING AI?

AI has become a practical companion rather than a replacement, helping structure plans, speed up analysis, automate workflows, and present findings more engagingly. It streamlines tasks, generates guided insights, and boosts company-wide engagement while still grounding everything in real user data. The result is faster research cycles that free up time for the human essentials of the job—judgment, creativity, and empathy.

PROMPT DESIGN FOR ARTISTS: HOW TO GET BETTER RESULTS FROM AI TOOLS

AI-generated art tools like Midjourney and Leonardo AI transform text into images within seconds.