Search the archive

GLASS, NOT ICE

Cryopreservation is a technology that brings tissue into a temporary glassy state to pause biological time so that patients can have more time. It is used in IVF and to help people recover from chemotherapy through hematopoietic stem cell transplants. The technology uses controlled-rate freezers, perfusion systems, novel cryoprotectants, and rewarming devices to preserve complex biological systems. The field is aiming for whole-body medical hibernation, but there are many fundamental questions that must be answered before we can start heading toward that goal.

agents fail to deliver dependable results. This guide from You.com

breaks down the evolution of agent management, revealing the five stages for building a successful AI agent and why most organizations haven't gotten there yet.

you.

HL

hl is a high-performance log viewer and processor that transforms logs in JSON and logfmt formats into a human-readable output. It enables quick parsing and analysis of large log files with minimal overhead. hl features automatic pager integration, field-based filtering, level filtering, time-stamp range filtering, field visibility control, and more. It currently doesn't provide a built-in pager - users are recommended to install 'less'.

HOW TO WRITE A GREAT AGENTS.MD: LESSONS FROM OVER 2,500 REPOSITORIES

GitHub's agents.md feature allows developers to create agents with custom instructions. agents.md is where developers define all of the specifics, provide code style examples, and set clear boundaries of what not to do. Most agent files fail because they're too vague. The best files provide agents with a clear job or persona, exact commands to run, well-defined boundaries to follow, and clear examples of good output for the agent to follow.

FROM NETWORK EFFECTS TO COGNITIVE EFFECTS: THE NEW RULES FOR PLATFORM DOMINANCE

Cognitive design stitches memory, intuition, and prediction into something that feels like a partnership. The best companies design for anticipation. Mastering this balance results in products that feel simultaneously reliable and almost alive. The future belongs to companies that understand their users deeply and produce products that already know what customers need.

PRESENTATIONS

Benedict Evans produces a presentation exploring the macro and strategic trends in the tech industry twice a year. The most recent presentation is titled 'AI eats the world'. This post contains the slides for that presentation. It also includes the videos for Evans' presentations at SuperAI in Singapore in Spring 2025 and Slush in Helsinki in November 2024.

THE CHIP CEO STARING DOWN NVIDIA AND TALK OF AN AI BUBBLE

Lisa Su, AMD's CEO, recently announced she was pivoting the entire company, putting AI at the center of its entire product line.

OVER-REGULATION IS DOUBLING THE COST

While basic and strong protections for clear harms are needed, we need to allow people to invent and build hardware again.

CHATGPT LAUNCHES GROUP CHATS GLOBALLY

ChatGPT group chats, which allow users to collaborate with each other and ChatGPT in one shared conversation, are now available globally to all users on Free, Go, Plus, and Pro plans.

TESLA DEPLOYS ITS FIRST SUPERCHARGERS OWNED BY 3RD-PARTY IN THE US

Suncoast Charging in Land O'Lakes, Florida, is owned by the host but fully managed by Tesla.

THE INTEGRAL OF INTENTIONALITY

Humans subconsciously tally up human effort when looking at things.

BROWNOUTS REVEAL SYSTEM BOUNDARIES

You can't control what fails, but you can control whether you make something a single point of failure.

Download the free ebook

GEMINI 3 PRO IMAGE

Nano Banana Pro, aka Gemini 3 Pro Image, claims studio-quality image gen and editing. It can render text (a historic struggle), consistent multi-character scenes, and real-world knowledge integration for creating infographics.

A NEW WAY TO COLLABORATE IN CHATGPT

Group chats are now rolling out globally to ChatGPT Free, Go, Plus, and Pro plans. Users can now invite up to 20 other people into a shared conversation with ChatGPT. Group chats are optional and separate from private conversations. Personal ChatGPT memory isn't shared in group chats, and no new memories are created from them. Users can only join group chats via invitation, and they can choose to leave or remove members from the chat at any time.

SOME THOUGHTS ON AI

Gemini 3 shows that scaling laws for pretraining are intact. This means that Blackwell models are likely to show a significant increase in performance when they come out next year. GPT-5 was not evidence of a slowdown in scaling laws - the model was designed to be cheaper to inference, not better. While Gemini 3 was trained on TPUs, coherent FLOPs is what matters for pre-training, so the results should transfer to Blackwell chips fine.

OPENAI CAN'T BEAT GOOGLE IN CONSUMER AI

OpenAI can't beat Google at consumer AI as long as it is still in the chatbot paradigm. Recent OpenAI launches haven't done well. OpenAI needs new form factors to drive incremental growth. Google's data advantage is really showing, and its little advantages from its wide variety of apps add up. Microsoft and Amazon are at risk of losing the largest AI workloads.

THE SCALING WALL WAS A MIRAGE

Gemini 3 showed that models can still improve by adding more compute during scaling. It has the same parameter count as Gemini 2.5 but with massive performance improvements. Google credits the improvements to pre-training and post-training and says that there is still no limit in sight. The model release is the strongest evidence since OpenAI's o1 that pre-training scaling still works when algorithmic improvements meet better compute.

PARALLEL EXTRACT

Parallel Extract is an API that fetches content from URLs and returns it as Markdown. It is built on the same proprietary web index and retrieval infrastructure that powers Parallel's Search, Task, FindAll, and Monitor APIs. Parallel Extract can reliably extract information from the most challenging sites. Developers can try Extract for free in the Parallel Developer Platform.

INTRODUCING CLINE-BENCH: A REAL-WORLD, OPEN SOURCE BENCHMARK FOR AGENTIC CODING

cline-bench is a new initiative focused on creating high fidelity benchmarks and reinforcement learning environments derived from real open source development scenarios. AI models have advanced significantly, but the field still lacks a rigorous open source benchmark that represents real engineering work. Model labs require evals that expose real breakdowns. cline-bench supports the next stage of AI research and development by creating research-grade environments that capture actual engineering constraints.

AUTOCOMP

Autocomp is a framework for optimizing tensor accelerator code. It helps hardware designers extract the full performance of tensor accelerators. Autocomp outperforms human expert kernel writers by up to 17x on AWS Trainium. It is highly portable and easy to use.

ACT-1

ACT-1 is a frontier robot foundation model trained on zero robot data. It was trained with special Skill Capture Gloves, which allow for data collection from anywhere without having to move robots around. The glove-based data collection produces higher quality data than teleoperations on contact-rich tasks. Videos of Sunday Robotics' Memo robot performing tasks using the ACT-1 model are available in the thread.

OPENAI'S DOMINANCE IS UNLIKE ANYTHING SILICON VALLEY HAS EVER SEEN

OpenAI dominates the AI landscape with secretive financials and a rapid expansion strategy across data centers, applications, and devices. CEO Sam Altman's aggressive partnerships with Nvidia, Broadcom, Oracle, and AMD underscore a $500 billion growth trajectory. Despite concerns about competition, industry leaders see OpenAI's rapid pace as both a challenge and an asset, sparking a "gold rush mentality" in AI investments.

AI IMAGE VERIFICATION IN GEMINI

Google has integrated SynthID-based AI image verification into the Gemini app, allowing users to check whether an image was generated or edited by Google AI directly through simple queries.

EARLY EXPERIMENTS IN ACCELERATING SCIENCE WITH GPT-5

OpenAI published case studies from collaborations with institutions, including Oxford, Berkeley, and Lawrence Livermore, showing GPT-5 helping researchers solve problems that took months in minutes.

FOXCONN, OPENAI PARTNER ON AI HARDWARE MANUFACTURING

In the latest in a string of high-profile deals, OpenAI will work with Foxconn on the design of data center racks and AI hardware components.

AI2'S OLMO3 TECHNICAL REPORT

Olmo 3 is a new suite of fully open models at 7B and 32B sizes.

BANKS' CRYPTO RULES MUST BE REWORKED

Basel Committee chair Erik Thedéen acknowledged crypto capital rules need reworking after the US and UK refused to implement the 2021 framework requiring banks to hold large capital reserves against crypto losses, with Fed Vice-Chair Michelle Bowman calling the rules "not very realistic." The original framework focused on volatile assets like Bitcoin but failed to account for stablecoins' rapid growth and lower risk profiles. The Bank of England declined to implement the rules in their current form as the committee faces pressure to quickly reassess whether stablecoins warrant the same capital treatment as traditional cryptocurrencies ahead of the framework's planned January implementation.

BITCOIN HOLDS AMID SELL PRESSURE

Bitcoin absorbed $228 million in long-term holder sales as trading volume rose 5% to $81 billion, while over $600 million in leveraged futures liquidated primarily from unwinding bullish positions. Altcoins declined with ETH down 3.4% and the altcoin season index falling to 26/100, though Cosmos surged 10% on a technical breakout and Zcash extended its two-month rally with an 8.7% gain. Memecoins are underperforming the rest of the market.

AZTEC IGNITION CHAIN IS LIVE

Aztec launched Ignition Chain on the Ethereum mainnet, positioning it as the first fully decentralized L2 after the validator queue hit 500, triggering block production with 525 validators now running across 125 nodes. Staking requires 200,000 AZTEC tokens per validator node and currently offers 167% APY as early sequencer rewards accelerate decentralization. The AZTEC token sale launches on December 2 with a $350 million starting floor that increases with demand, featuring early access for select solo staker communities, with tokens used for staking, governance, and block rewards.

HELLOTRADE: FRICTIONLESS ACCESS TO GLOBAL EQUITIES

HelloTrade was founded by former BlackRock crypto directors Wyatt Raich and Kevin Tang, who launched IBIT, the fastest ETF to $100 billion in assets. Built on MegaETH's 100,000+ TPS infrastructure, the mobile-first platform will offer leveraged exposure to global stocks, ETFs, commodities, and crypto without wallet setup or gas payments to remove geography and capital requirement barriers. The iOS and Android apps will launch soon with an active waitlist at hello.trade.

STABLECOIN CARDS VS BANKS

Today's stablecoin cards ride on Visa and issuing banks' infrastructure, taking a slice of interchange while offering better UX, global access, DeFi yield on idle balances, and token-subsidized cashback, which makes them look insanely attractive but effectively eats into banks' margins. This won't last forever because, as crypto cards start replacing primary bank accounts rather than adding extra volume, banks and regulators will fight back by squeezing interchange and reclaiming economics, forcing stablecoin card players either to own more of the payment stack (licenses, rails, maybe bank-to-bank or wallet-to-wallet) or accept a hard cap on their upside.

THE GREAT CONSOLIDATION

The Great Consolidation is accelerating as $70 billion in altcoin unlocks approach and fragmented liquidity across 25,000+ daily token launches becomes unsustainable, with most alts down 50-70% from highs as Bitcoin dominance rises and institutional flows bypass alt season entirely. The industry must consolidate memes toward lindy names, eliminate copycat Alt-L1 ecosystems offering no unique value proposition, and push apps toward superapp models as solo farming becomes unprofitable and retail rejects low float, high FDV exit liquidity games. Ethereum faces existential challenges, capturing only 1% of Base fees while ETH/BTC hits multi-year lows with no clear value accrual narrative as L2 fragmentation and stablecoin adoption undermine the moneyness thesis, though blobs and potential L2 "tariffs" offer long-term scalability paths. Winners will be fee-generating protocols like Aave, Ethena, Uniswap, and Hyperliquid that established product-market fit before token launch as fundamentals replace speculation and supply overhangs crush hype-driven projects.

THE REAL SUPERCYCLE

The “supercycle” is happening quietly in fundamentals, not prices. Most speculative tokens and 2021 narratives are dying, while critical infrastructure like Ethereum/Solana, stablecoins, RWAs, and DeFi protocols with real fees keep making new highs in usage and TVL. Regulatory clarity plus integrations from Visa, Mastercard, Stripe, neobanks, and apps like Cash App and Revolut are wiring stablecoins into global payments, and as rates fall, capital should rotate from Treasuries into onchain, fee-based yield. The next leg comes from AI agents using blockchains as their native financial backend, meaning we've already hit escape velocity even if 90% of legacy coins never recover.

DECENTRALIZATION OR BUST

Today's “dead” crypto vibe is less about price and more about the industry forgetting its purpose. BTC is increasingly treated as a Wall Street ETF game. Much of crypto has been reduced to casino tokens rather than tools for censorship resistance and permissionless value transfer. If assets are custodial and centralized, they become as seizable as any bank IOU and drift toward ponzi-like dynamics where new buyers don't strengthen the network, only transfer wealth to early holders. Refocus relentlessly on decentralization and self-custody, avoid perps and degen trading with most of your net worth, hold a core stack of BTC/ETH/NFTs you truly value, keep working in the real world, and remember that most coins and JPEGs will go to zero, but the long game of decentralized systems is still worth playing.

ETHEREUM INTEROPERABILITY LAYER TO MAKE L2S FEEL LIKE ONE CHAIN

Ethereum's Interoperability Layer, set to launch in 2026, will let users sign a single transaction while cross-chain paymasters and liquidity providers handle gas and settlement across multiple L2s behind the scenes, making all Ethereum L2s feel like one unified chain without relying on trusted middlemen.

BLACKROCK LINES UP STAKED ETHEREUM ETF

BlackRock is preparing a second ETH ETF focused on staking, likely staking most of the fund for higher real yield while only partially staking assets in its original ETH product.

ETH STAKING HITS NEW ALL-TIME HIGH

Despite choppy price action and bearish sentiment, the amount of ETH staked has hit a new all-time high of over 33 million.

BEAR MARKET READING LIST: CRYPTO WHITEPAPER TIER LIST

Folks looking for bear market reading should begin with the Bitcoin, Ethereum, Uniswap v2/v3, Zerocash, Augur, Flash Boys 2.0, and Truebit whitepapers, followed by Avalanche, Cosmos, MakerDAO, EigenLayer, and some others.

FIGMA INTRODUCES MAKE CONNECTORS FOR EXTERNAL INTEGRATIONS

Figma's Make Connectors enables direct integration of external tools like Notion, Asana, Linear, GitHub, and Jira into Figma Make via the Model Context Protocol. The feature allows teams to import documents and tasks, refresh data in real time, and create or modify records in connected apps without leaving Figma, with granular access controls and user approval required for write operations. Make Connectors are available on paid plans.

LUMA AI RAISES $900 MILLION FROM SAUDI-BACKED HUMAIN

Luma AI has secured $900 million in funding, led by Saudi-backed HUMAIN, bringing total funding to over $1 billion at a reported $4 billion valuation. The startup, known for Dream Machine and the Ray model, will use the capital to advance large-scale World Models and become one of the first customers of HUMAIN's two-gigawatt Project Halo supercluster in Saudi Arabia. Additional investors include AMD, Amazon, Andreessen Horowitz, Amplify Partners, and Matrix Partners. The funding was announced during Saudi Crown Prince Mohammad bin Salman's US state visit.

CHATGPT LAUNCHES GROUP CHATS GLOBALLY

ChatGPT has rolled out group chats worldwide, letting up to 20 people collaborate with each other and the assistant in a shared conversation. The feature is designed to help users plan, create, research, and make decisions together while keeping individual settings private. OpenAI sees this as a step toward making ChatGPT a more social and collaborative platform rather than a purely one-on-one tool.

TO GROW, WE MUST FORGET… BUT NOW AI REMEMBERS EVERYTHING

AI systems with infinite memory risk trapping users in outdated versions of themselves by constantly reinforcing past preferences and behaviors. Human forgetting evolved as a survival advantage that enables adaptation, creativity, and growth, but memory-enabled AI creates echo chambers that stagnate cognitive development by optimizing for comfort over challenge. The solution requires action at three levels: users must actively manage their stored data, UX designers should build flexible "sketchpad" memory interfaces with expiration dates and revision tools, and developers should engineer adaptive forgetting systems that mimic the brain's natural memory processes.

AN ALTERNATIVE CHAT UI LAYOUT

This post proposes a dual-scroll pane layout for AI chat interfaces to address usability issues posed by agentic features such as thinking traces and tool calls. The design separates user instructions and AI process in one column from results in another, keeping both in context without excessive scrolling. This approach aims to solve the problem of long AI monologues pushing initial messages off-screen, particularly in narrow side panels.