Search the archive

THE UNPOWERED SSDS IN YOUR DRAWER ARE SLOWLY LOSING YOUR DATA

SSDs can suffer from corrupted or lost data if left unpowered for extended periods.

PEBBLE WATCH SOFTWARE IS NOW 100% OPEN SOURCE + TICK TALK #4 - PT2 DEMOS!

Pebble owners now have all they need to build, run, and improve the software behind their devices.

GOOGLE REVISITS JPEG XL IN CHROMIUM AFTER EARLIER REMOVAL

JPEG XL is a replacement for JPEG that introduces animation support and also reduces file size without a loss in visual quality.

META, GOOGLE DISCUSS TPU DEAL AS GOOGLE TARGETS NVIDIA'S LEAD

Meta is reportedly in discussions to spend billions to integrate Google's TPUs into its data centers starting in 2027 and also rent TPU capacity from Google Cloud.

Register now

INTRODUCING CLAUDE OPUS 4.5

Claude Opus 4.5 is the first model to exceed 80% on SWE-bench Verified. It achieves state-of-the-art results across coding, tool use, and reasoning benchmarks. The model is priced at $5/$25 per million tokens, down from previous Opus pricing. It includes a new "effort" parameter letting developers trade speed for capability, automatic context compression enabling unlimited conversation length, and expanded availability for Claude for Chrome and Claude for Excel.

INTRODUCING SHOPPING RESEARCH IN CHATGPT

Just in time for the holidays, OpenAI launched an interactive product discovery feature that searches the web, asks clarifying questions, and generates personalized buyer's guides. Users can guide research in real-time by marking products "Not interested" or "More like this." The system draws on ChatGPT memory for personalization. It's powered by a GPT-5 mini model trained specifically on shopping tasks.

NANO BANANA-PILLED, NANO BANANA-SHILLED FOR SPACESHIP ENGINEERING

Nano Banana Pro can make good diagrams based on papers. It can also make pretty good presentations, even on the free tier. This post presents some examples of what happens when you feed the model essays on spaceship engineering. While the facts within the slides still need to be verified, it is clear that the technology is trending towards something impressive - it can only get better from here.

A TSUNAMI OF COGS

The AI industry is in correction mode. OpenAI, Anthropic, and Cursor are subsidizing demand with negative margins. Google was surprised by the AI boom, and it took a while for it to get its act together, but it is now coming back strong. Its pockets are full, and it is better positioned to play the negative margin game. If challengers don't want to drown in a tsunami of costs, something has to change.

UNIVERSAL LLM MEMORY DOES NOT EXIST

Semantic memory tracks preferences, long-term history, and rapport. Working memory tracks file paths, variable names, and immediate error logs. Semantic memory is brilliant for personalization across sessions, but bad for execution state within a task. Treat semantic memory and working memory as separate systems with separate requirements.

Download the guide.

INTRODUCING ADVANCED TOOL USE ON THE CLAUDE DEVELOPER PLATFORM

Anthropic released three beta features for developers. Tool Search Tool discovers tools on demand rather than loading all definitions upfront, reducing token consumption by 85%. Programmatic Tool Calling lets Claude orchestrate multiple tools through Python code instead of individual API calls, cutting token usage 37%. Tool Use Examples provides concrete usage patterns beyond JSON schemas, improving accuracy from 72%-90% on complex parameter handling.

BUILDING AN AI-NATIVE ENGINEERING TEAM

AI coding agents are revolutionizing the software development lifecycle by managing tasks from scoping and prototyping to implementation and operational triage, allowing engineers to focus on architecture and product intent. These agents now sustain multi-hour reasoning, effectively contributing across planning, design, development, testing, code reviews, and deployment. Teams that adopt coding agents for well-defined tasks can achieve faster delivery and improved efficiency without drastically altering existing workflows.

TAKING JAGGEDNESS SERIOUSLY

Uneven capability advances in AI will persist because some tasks have clear, verifiable rewards that can be used in reinforcement learning, but most real work doesn't. Most work requires gathering and synthesizing information across different systems and human relationships. Organizations redesigning workflows around AI's strongest current capabilities will gain power over those waiting for the industry's promised "drop-in remote worker" that can handle everything.

AI MEETS AGGRESSIVE ACCOUNTING AT META'S GIGANTIC NEW DATA CENTER

Meta is building a $27 billion data center financed with debt. Neither the data center nor the debt will be on its own balance sheet. Meta will rent the data center for up to 20 years, beginning in 2029, starting with a four-year lease term with options to renew every four years. The lease structure minimizes the lease liabilities and related assets that Meta will recognize.

IS AI REALLY EATING THE WORLD?

The current evidence points towards commoditization, as value is flowing up the stack rather than to model providers.

THE BITTER LESSON OF LLM EXTENSIONS

We've gone from simple system prompts to complex client-server protocols and back again.

JONY IVE AND SAM ALTMAN SAY THEY FINALLY HAVE AN AI HARDWARE PROTOTYPE

OpenAI's first hardware device could arrive in under two years.

MEET THE NEW CHINESE VIBE CODING APP THAT'S SO POPULAR, ONE OF ITS TOOLS CRASHED

Ant Group's LingGuang vibe coding app surged to over 2 million downloads in six days (hitting its first million faster than ChatGPT or Sora).

AI ASSISTANCE IS NOW BECOMING ROUTINE OVER AT THE ERDOS PROBLEM WEBSITE

Top mathematicians are using AI to generate proofs and solve problems.

$AMZN AWS RAN INTO AN AI CAPACITY CRUNCH THIS SUMMER

AWS Bedrock hit critical capacity constraints, causing some customers to move to rivals like Google Cloud.

BITCOIN BOUNCE, STILL FRAGILE

Bitcoin has rebounded to around $87.5K after dipping near $81K, with analysts calling it a “post-flush bounce” rather than a full recovery. Market structure is fragile. BTC is expected to chop in a tight $85K–$90K range while long-term holders quietly accumulate.

BITCOIN LOGS LONGEST LOSING STREAK SINCE 2024

Bitcoin tracked its fourth consecutive weekly loss, marking the longest downtrend since June 2024. Q4 2025 looked like its worst quarterly performance since 2018 at -24% despite bouncing 6% from the November 21 low of $82,100 to $87,400. The aggregate spot bid-ask delta at 10% depth spiked to 2025's second-highest level, indicating dip-buying activity, while Federal Reserve rate cut odds surged from 40% to 70% last week. There is a potential bull trap with options traders loading puts in the $80,000-$85,000 range, meaning Bitcoin could briefly slip to mid-to-high $70,000s before recovering to $90,000 by year-end if the Fed avoids hawkish tone. Recovery to $100,000 is expected by Q1 2026.

REAL-TIME BLOCK PROVING USING AIRBENDER

zkSync's Airbender prover achieved real-time Ethereum L1 block proof generation using just two consumer-grade RTX 5090 GPUs, prompting Ethereum Foundation researcher Justin Drake to suggest the breakthrough could enable "gigagas L1" with significantly higher gas limits pushing L2 fees toward zero. Vitalik Buterin praised the "huge milestone" but cautioned about worst-case performance gaps where proving times can differ by 60x between average and worst-case blocks, recommending repricing gas for expensive operations like RSA verification rather than hardware arms races. The achievement was showcased at Ethproofs Day in Buenos Aires and marks a shift in Ethereum's scaling trajectory as consumer-grade proving becomes feasible compared to historical data center resource requirements.

LEVERAGE ON PREDICTION MARKETS IS POSSIBLE

Leverage in prediction markets faces fundamental challenges because instant price jumps to zero eliminate liquidation windows, requiring financiers to price expected losses upfront, but naive probability-based fees merely create accounting tricks where levered ROI equals unleveraged positions. The solution splits price movements into "creep" (continuous drift allowing liquidation) versus "jumps" (instant gaps with no reaction time) and prices risk over short epochs with rolling per-epoch fees similar to perps funding rather than pricing entire position lifespans. This epoch-based approach treats drift, volatility, and jump rates as frozen constants within small time windows, computing liquidation probabilities and average losses separately for creep versus jumps, while limiting misspecification risk that could wildly over- or underestimate fees as market conditions shift near resolution dates. The architecture includes block-level auction designs capturing news-driven arbitrage to rebate market makers and financiers, further reducing effective jump losses and fees charged to traders.

HOW TO PROFIT ON CORRELATED POLYMARKET BETS

You can often get better risk/reward on Polymarket by trading relationships between markets instead of single yes/no outcomes. Three ways to do this include time spreads on the same event, “either-extreme-wins” baskets, and logical/arbitrage-style bets across correlated AI markets.

$12B IN DEFI LIQUIDITY SITS IDLE

83-95% of liquidity in major DeFi pools, including Uniswap and Curve, sits idle, with only 0.5% of Uniswap v2's liquidity falling within active trading ranges, leaving nearly $1.8 billion ineffective. Retail liquidity providers bear the brunt as 50% lose money from impermanent loss with net deficits exceeding $60 million, including one Uniswap v3 pool losing over $30 million to Just-in-Time manipulation. 1inch's Aqua protocol could be a solution as it allows DeFi applications to share a common capital base across strategies while users keep assets in wallets and create "virtual trading positions," with any existing DEX implementable under 10 lines of code. The liquidity crisis stems from over 7 million fragmented pools across the ecosystem diluting capital and reducing routing efficiency for trades.

PREDICTION MARKET FEE WARS BEGIN

Big brokerages in the US are starting to treat prediction markets as commoditized backends. Kalshi is already rebating up to half its fees to Webull for order flow, while Robinhood is signaling it will route prediction orders across multiple venues and eventually squeeze fees or build its own markets.

OFFCHAIN LABS CHALLENGES VITALIK'S RISC-V PROPOSAL

Offchain Labs challenged Vitalik Buterin's proposal to transition Ethereum's execution layer to RISC-V, arguing WebAssembly offers superior long-term advantages by distinguishing between "delivery ISA" for uploading contracts versus "proving ISA" for ZK virtual machines. Arbitrum already proves WASM-based Stylus smart contracts by compiling WASM to RISC-V first, and enshrining RISC-V on L1 could lock Ethereum into proving technology just as better alternatives emerge. WASM's structured design, efficient execution on common hardware without emulation, validation capabilities preventing vulnerabilities, and mature tooling ecosystem position it as an "Internet Protocol for smart contracts" as ZK proving costs have plummeted to $0.025 per block.

EIGENDA'S “LITTLE GREEN SLICE”

Ethereum's data chart shows L1 blobs still dominate rollup data (≈67%), with EigenDA already at ~19% and Celestia the rest.

BITCOIN VOLATILITY VS. ETHEREUM SETTLEMENT

Crypto feels like it's hanging by a thread as BTC has dropped from ~126K to ~85K in weeks, wiping out around $1T in market cap.

GIP-143: GNOSISDAO FIRES KARPATKEY TREASURY MANAGER

GnosisDAO has voted overwhelmingly to end Karpatkey's mandate to manage its ~$300M non-GNO treasury.

JPMORGAN SHUTS STRIKE CEO'S ACCOUNTS

Strike CEO Jack Mallers says JPMorgan Chase abruptly closed his personal bank accounts.

INTRODUCING REPLIT DESIGN MODE

Replit's Design Mode is powered by Google's new Gemini 3 model. It enables users to create interactive mockups and static websites in under two minutes through natural language commands. The feature serves both as a rapid prototyping tool for designers and product managers, and as a faster alternative to website builders like Wix or Squarespace for creating deployable sites. Users can convert their front-end designs into full-stack applications with backend functionality by clicking "Convert to App." The entire development process happens within Replit's workspace.

GOOGLE AI MODE IS STILL TESTING SHOWING ADS IN SEARCH RESULTS

Google is testing ads in its Gemini-powered AI Mode search results, placing sponsored cards at the bottom of pages rather than at the top like traditional searches. The company confirmed these intermittent ads are part of ongoing tests announced in May, with no new updates to AI Mode's ad distribution currently. Organic link cards now appear within Gemini's answer above the ads, and unlike traditional search, there's no option to hide sponsored results in AI Mode.

AI TOOK CENTRE STAGE AT ADOBE MAX LA 2025

Adobe MAX LA 2025 showcased Adobe's push to make AI a powerful assistant rather than a replacement, integrating conversational agents across Photoshop, Firefly, and Express to handle repetitive tasks through plain-language commands. Major updates included the advanced Firefly Image Model 5, refined generative tools in Creative Cloud, new video features like AI Object Mask, and enterprise solutions such as GenStudio and Firefly Foundry—aimed at speeding up workflows while keeping creative control firmly with humans.

GENERATIVE UI FROM GEMINI 3 PRO

Google's Gemini 3 Pro introduces Generative UI, where AI creates custom interfaces on the fly for individual users. Test users preferred these over traditional websites 90% of the time. Human designers currently edge out AI designs by just 56% to 43%, but this narrow margin is expected to disappear by late 2026 as AI capabilities double roughly every seven months while human design skills stagnate. Google plans to roll out Generative UI to its five billion search users, marking a shift from expensive, permanent interfaces toward cheap, disposable, individualized interactions generated instantly for specific user needs.

HUMAN ENGINEERING

Human engineering is a multidisciplinary field focused on designing systems, technologies, and environments that make correct actions intuitive and errors difficult, rather than relying on training or willpower. This article illustrates this through everyday examples like stovetop configurations and "Norman Doors," demonstrating how poor spatial compatibility and discoverability lead to common mistakes that stem from design failures rather than human error. By applying principles such as natural mapping, affordances, and constraints, human engineering reduces cognitive load and creates systems where ordinary human performance produces excellent results at minimal cost.

QUALITY OF DESIGN EDUCATION "LOWER THAN IT WAS" SAY DESIGNERS

Design education has deteriorated: it's more expensive, less connected to industry, and leaves graduates unprepared and in debt. Too many under-resourced courses lack real-world experience, sustainability training, and adequately supported teachers, suggesting that many skills can now be self-taught and that the traditional university model may need to be rebuilt entirely.

BETTER MARKETING DECISIONS 10X FASTER

GoMarble is an AI agent that analyzes your paid marketing and gives instant insights, creative analyses, and expert recommendations.

FREE ONLINE IMAGE CONVERTER

Convert images between formats, locally in your browser, no uploads, 100% securely and privately.

12 TOP DESIGNERS TELL US HOW THEY MADE IT

Twelve established designers, including Philippe Starck and Yinka Ilori, candidly share how they broke into the industry, revealing stories of patience, side jobs, and serendipitous connections that launched their careers. Many express concern about today's design landscape, citing reduced funding, expensive trade shows, and inadequate education as major barriers for emerging designers compared to when they started. Despite these challenges, the designers note that social media and independent approaches have created new pathways, with Instagram facilitating collaborations and designers increasingly bypassing traditional trade shows to take control of their own work presentation.

WHEN NATURAL LANGUAGE FEELS UNNATURAL

ChatGPT popularized natural language as a consumer interface in late 2023, but chat's dominance as the primary AI interaction method may be temporary, much like command-line interfaces before graphical user interfaces emerged. While natural language excels for knowledge work and coding, where ambiguity is tolerable, it struggles with precision tasks, leading to unwanted rewrites and misinterpretations that reveal chat as an incomplete interface layer. The field awaits its "Mother of All Demos" moment—a breakthrough that integrates multimodal inputs, contextual affordances, and dynamic interfaces into a new interaction grammar beyond simple text prompts.

NINETY YEARS OF PENGUIN: HOW TYPOGRAPHY BUILT A PUBLISHING ICON

Penguin's identity has always been rooted in typography, evolving from early consistent paperbacks to Jan Tschichold's 1940s philosophy of clarity, structure, and reader-centered design. Across decades—from Gill Sans modernism to Helvetica's international style to later post-modern experimentation—Penguin has adapted to cultural and technological change while preserving its core values of clarity, consistency, and respectful communication, showing how typography can shape both reading and culture itself.

WHY 85% OF AI PROJECTS FAIL (AND HOW UX CAN PREVENT THESE COSTLY MISTAKES)

85% of AI projects fail primarily because they lack valid use cases that address real user needs rather than just business objectives.

LONDON'S NIGHTMARISH CHRISTMAS MURAL SHOWS WHAT HAPPENS WHEN AI REPLACES DESIGNERS

A London developer used generative AI to create a Christmas banner for a Kingston-upon-Thames development, resulting in a disturbing mural featuring distorted faces, malformed snowmen, and chimeric creatures that locals compared to "Lovecraftian horror."

THIS FOOTBALL TEAM'S NEW LOGO IS PURE FAN SERVICE (AND I'M NOT MAD)

Iowa State has revived its beloved “Cynado” tornado-mascot logo with a modern refresh, delighting fans who hope it replaces the unpopular I-State wordmark permanently.