Search the archive

A HACKING TOOL BUILT WITH AI CAN BREACH PHONES WITHOUT A CLICK

WeWorm is a zero-click attack that can compromise WeChat accounts without users doing anything and gain access to messages, calls, and accounts.

GOD HELP US, LET'S TRY TO LEARN ABOUT MECHANISTIC INTERPRETABILITY TECHNIQUES

Mechanistic interpretability is the science of reading an AI's mind.

APPLE ACQUIRES STARTUP WORKING ON 'BREAKTHROUGH SENSING TECHNOLOGY'

Sonera is working on sensing technology that non-invasively measures magnetic fields generated by the brain and body to analyze neural data without direct skin contact.

GOOGLE MAPPED A FRUIT FLY'S BRAIN. NOW IT'S PLAYING DOOM AND SUPER MARIO 64

Google's fly connectome contains over 166,000 neurons and 125 million synaptic connections, providing scientists with a fundamental resource for studying how the brain works.

Get the guide.

CHATGPT IMAGES 2.5

OpenAI introduced ChatGPT Images 2.5 with sharper details, better reference-image preservation, more reliable editing, and up to 50% lower generation latency.

AN OPENAI MODEL SOLVED THE NAVIER–STOKES MILLENNIUM PROBLEM

OpenAI announced that an internal AI system produced a proof resolving the roughly 90-year-old Navier–Stokes existence and smoothness problem, one of mathematics' seven Millennium Prize Problems. The model showed that smooth three-dimensional fluid dynamics can develop a finite-time singularity and produced both an analytical proof and a Lean formalization.

INTRODUCING MUSE: THE WORLD'S FIRST PERSONAL AI AGENT BUILT FOR EVERYONE

Meta introduces Muse, a personal AI agent, powered by Muse Spark, to help users achieve goals by automating tasks like booking travel or sending emails. Muse operates securely on Muse Secure VM, ensuring data privacy with unique protections like the Sentinel agent overseeing actions. Muse will soon offer encrypted data with Muse Confidential VM and is available on iOS, Android, and muse.ai in the US.

>10X MORE EFFICIENT PRETRAINING

Without large amounts of compute, small labs can only compete through algorithmic efficiency. Magic's pretraining recipe is now more than 10 times more compute-efficient than that of leading open-weight base models. The startup believes that pretraining, agentic RL, and long-context are sufficient for building superhuman coding agents and automating AI research and development. This post discusses its pretraining and long-context work.

INSIDE THE MEGAKERNEL SERVING ENGINE FOR NORTH MINI CODE

This post presents a fully fledged serving system built around a decode megakernel. The system supports everything a real server needs: continuous batching, paged attention, and ragged sequence lengths, all behind an OpenAI-compatible endpoint with tool calling. The megakernel reaches 292 tokens per second on batch size 1, or 62% of SoL - 1.58× faster than vLLM. That margin holds across batch sizes and out to 256K of context with no measurable loss of accuracy.

PRETRAINING PROGRESS IS MOSTLY COMING FROM DATA

Between 2019 and 2025, 3.24x more compute efficiency gains have come from data improvements rather than model improvements. The gains from data and model improvements are mostly independent and don't interact. Most model research has consisted of removing or pushing back constraints to scaling. The data improvements may matter less for larger models. Small models see significant gains from data quality.

INTRODUCING MERCURY 2.5

Mercury 2.5 is the largest diffusion language model ever trained. It performs comparably to cost-optimized frontier models like GPT-5.6 Luna (Low), Gemini 3.5 Flash-Lite, and Claude Haiku 4.5. The model outputs 1,107 tokens per second on widely available Nvidia GPUs and has a 260K-token context window. At launch, Mercury 2.5 is 80% off at $0.04 per million input and $0.15 per million output.

HYPER-𝜏-BENCH: EVALUATING AGENTS THAT BUILD AGENTS

Hyper-𝜏-bench places a developer agent into a sandboxed workspace with the records of a simulated business and a simulated client that it can message at any time. The developer agent recovers the spec from the evidence, designs the architecture, and turns the business' actions into tools until it has a working customer-service agent. The finished agent has to serve from a fixed menu of models within a cost budget per conversation. Claude Opus 5 (max reasoning) running in Claude Code passes just 23.9% of the held-out evaluation tasks when working alone. Paired with an engineer with deep context, the same class of model reaches 82.2% on the same tasks.

GOOGLE'S ALPHAGENOME MAPS 9 BILLION GENETIC VARIANTS

Google DeepMind introduced AlphaGenome Atlas, a 1-petabyte database predicting the regulatory effects of all 9 billion possible single-nucleotide variants in the human genome.

COGNITION HITS $48B VALUATION, SIGNALING INVESTORS BELIEVE AI CODING IS FAR FROM A WINNER-TAKE-ALL MARKET

Cognition has raised $2 billion at a $48 billion valuation in a funding round led by Andreessen Horowitz, Accel, Founders Fund, General Catalyst, and Avenir. The startup's soaring valuation signals that VCs still see room for multiple major players to capture meaningful market share in AI coding. Cognition leases an Nvidia server cluster that could push its total cash burn to $800 million this year. The startup is expected to reach $4 billion to $5 billion in annualized revenue by the end of 2026.

I ASKED 100 AGENTS TO HACK ME

Around 100 self-hosted agents attempted to hack various online accounts over five hours. They compromised three accounts through software vulnerabilities and two through password brute-forcing, while also making 16 social engineering attempts. The experiment tested abliterated open-source models' capabilities, revealing potential future risks as these models improve and become cheaper to deploy.

IS THE 3X AI PRODUCTIVITY GAIN JUST A COMPUTER THAT NEVER SLEEPS?

OpenAI researchers now supervise 3.14 agent-workdays per eight-hour shift, suggesting AI productivity increasingly comes from parallel, around-the-clock machine labor rather than less human effort. That leverage is expensive, with median daily inference spend rising from $14 to over $600.

Apps. Tasks... Start free →

CHATGPT BROKE ITS MAU RECORD FOR THE 4TH CONSECUTIVE MONTH IN AUGUST

ChatGPT reached 1.06 billion monthly active users in August.

A RESPONSE TO BILL GATES'S ESSAY

Bill Gates' essay discusses AI's potential societal impact, proposing new institutions and taxes to manage transitions.

STEALING AI REASONING TRACES

It's possible to force a weaker, less safeguarded model from the same provider to decode and output reasoning traces verbatim in plaintext by injecting an encrypted reasoning trace from a target model without ever jailbreaking the more capable model directly.

ANTHROPIC RESEARCHER QUITS OVER ‘OUT-OF-CONTROL' AI FEARS

Jacob Coxon says he is leaving the company as he believes the industry-wide rush to build AI systems that can improve themselves could spiral out of control and destroy humanity.

PROGRESSIVE POINT MATCHING

Progressive Point Matching gives long-horizon RL partial credit without changing the optimal objective, improving training efficiency as tasks grow longer.

VISA'S STABLECOIN SETTLEMENT TOPS $20B RUN RATE

Visa's stablecoin settlement volume surpassed a $20B annualized run rate, up 15x year-over-year, with 160+ stablecoin-linked card programs live and payment volume on them up nearly 200%. Visa flagged a working-capital gap this creates for issuers (funding daily settlement before collecting from cardholders) and highlighted Credit Coop's stablecoin-denominated revolving credit facility, secured by settlement receivables and automated via smart contracts, which has cut borrowing costs up to 30% for participants. Rain has financed ~$2B through the facility with zero defaults since 2023, travel card issuer Karta (which also used Credit Coop) separately raised $140M in June.

ETHEREUM TARGETS QUANTUM-SAFE L1 BY 2029

The Ethereum Foundation published a tier list for the upcoming Hegotá upgrade, naming two "must-ship" proposals. They include FOCIL (EIP-7805), which strengthens censorship resistance by letting a validator committee force-include valid transactions, and Frame Transactions (EIP-8141), which makes account abstraction native by splitting transactions into programmable "frames". The Foundation set December 2029 as a firm target for full quantum resistance across execution, consensus, and data layers, treating that deadline as non-negotiable until it is reassessed in January, and is explicitly planning for the possibility of "Q-day" (when quantum computers could break current cryptography) arriving as early as 2030.

CRUMBS LAUNCHES STOCK-TOKEN SHOPPING REWARDS ON ROBINHOOD CHAIN

Crumbs, a pre-launch app on Robinhood Chain, converts verified retail purchase receipts into tokenized equity rewards from the shopped company, crediting Costco buyers with $COST tokens and Apple buyers with $AAPL. The model merges RWA tokenization with consumer loyalty mechanics, tying on-chain stock exposure to existing spending behavior rather than separate investment activity. The announcement drew 662 retweets and 1,956 likes before the official launch. Pre-registration is now open.

HUNTER BIDEN TO LAUNCH $LAPTOP MEMECOIN ON BASE ON SEPTEMBER 9

Hunter Biden is launching $LAPTOP on Coinbase's Base network on September 9. The token name references the 2020 laptop controversy. The tokenomics earmark 20% of the supply for investors who lost money on $TRUMP and include a conditional burn mechanism tied to 30 predetermined outcomes, including a Democrat winning the 2028 presidential election, Bitcoin reaching a new all-time high, and $LAPTOP's market cap surpassing Solana's. Up to 30% of total supply faces burns if all conditions are met, while missed targets redirect proportional token amounts to charity. The structure embeds outcome-based supply changes that place it closer to a prediction market instrument than a conventional memecoin.

HOW HEGOTÁ SHOULD APPROACH GAS REPRICING

Ethereum researchers have proposed a prioritization framework for gas repricing EIPs that targets the Hegotá fork, anchored by a unified data-floor pricing formula (EIPs 8131 and 8279) that produces a ~3.56% aggregate gas increase while pushing EIP-7702 transaction overhead from 0.04% to 20.63%. EIP-3298 eliminates storage-clear refunds to close GasToken arbitrage, and EIP-7923 replaces quadratic memory expansion with linear per-page charges plus a hard 64 MiB transaction cap, with both changes judged low-disruption relative to their scope. zkEVM-oriented EVMification changes (EIPs 7666, 8200) and BLOCKHASH repricing (EIP-7709) are recommended only if EIP-8025 ships concurrently. Whether Hegotá pursues the 450M gas limit increase determines the fork's scope: that scale target makes data repricing essential, while a narrower upgrade would defer contested proposals and concentrate effort on the highest impact-to-effort changes.

COLDCARD HACKER MOVES $7.7 MILLION IN BTC, 45% OF BITCOIN STOLEN IN THIRD ATTACK WAVE

The attacker behind the third wave of Coldcard hardware wallet thefts has moved $7.7 million in Bitcoin, representing 45% of the BTC taken during that phase, according to Galaxy Research. On-chain tracking confirms the attacker has fully drained all 11 of the largest vaults tied to the third wave, indicating a systematic approach to liquidating stolen holdings. The incident extends a serial campaign against Coldcard users across at least three documented attack waves, with each cycle followed by methodical fund movement that has so far evaded interdiction.

B2B OVER B2C FOR NEOBANKS

Neobanks should target businesses rather than consumers: the consumer product bar is low and crowded, while business banking's complexity is a real moat and generates far more volume per customer. Only a handful of players do it well globally, despite there being 350M businesses worldwide.

'THE CHAIN IS NOW EARNINGS': BERNSTEIN SEES UPSIDE FOR ROBINHOOD AS FEES TOP SOLANA, BNB CHAIN

Bernstein reiterated an Outperform rating on Robinhood Markets with a $160 price target (~31% upside), citing Robinhood Chain's ~$33 million in fees over a 15-day window, outpacing Solana (~$11 million) and BNB Chain (~$9 million) over the same period. Robinhood retains ~90% of collected fees, with Arbitrum receiving 10% for infrastructure and Ethereum under 1% for data availability, converting chain activity into a direct revenue stream on the company's income statement. Since the July 1 launch, the network has reached $1.5 billion in TVL, processed $50+ billion in DEX volume, grown stablecoin supply from $241 million to $1 billion, and expanded tokenized stock holdings 14x to $140 million, reflecting early traction across both DeFi and RWA verticals. Bernstein projects Robinhood Chain will reach ~$160 million in annualized fees by 2028.

BULL CASE FOR X402 AGENTIC PAYMENTS

x402 (a crypto micropayment protocol) removes friction for AI agents needing to pay for gated digital goods/services, making everything agent-discoverable and purchasable on demand.

BEAR CASE ON AGENTIC MICROPAYMENTS (UNDER 1 MINUTE READ)

The agentic micropayments narrative is mostly being pushed by innovation departments at large, established companies rather than startups or VCs, and shows little organic traction.

REDOTPAY RESPONDS TO FBI HAMAS AFFIDAVIT

RedotPay published a compliance update after an FBI affidavit referenced the platform in connection with Hamas, announcing a Big Four firm's independent review of its AML/CFT controls across its licensed Hong Kong subsidiaries.

SEC REPORTEDLY PREPARING TOKENIZATION INNOVATION EXEMPTION

An unconfirmed report suggests the SEC is preparing an innovation exemption that would permit tokenized securities to trade through transfer agents alone, eliminating broker-dealer licensing and exchange rule requirements for US retail and foreign investors.

👉 Launch your site with Framer today

CAMERA APP IN IOS 27 REPORTEDLY INCLUDES FOUR MAJOR PRO PHOTOGRAPHY FEATURES

Code discovered in the latest iOS 27 beta suggests Apple may bring advanced photography features to the Camera app, including exposure histograms, waveform monitors, focus peaking, zebra highlight warnings, and manual focus controls. The update could also add burst-style self-timer shots and image processing tailored for the rumored variable-aperture camera on the iPhone 18 Pro. If released, these features would significantly narrow the gap between Apple's built-in Camera app and professional third-party camera apps.

GEMINI OVERLAY GETS BUBBLE MINIMIZATION AND MULTITASKING ON ANDROID

Google is rolling out a new multitasking feature for the Gemini overlay on Android that lets users minimize conversations into a floating bubble for quick access while using other apps. The bubble can be repositioned to preset locations, although a bug currently resets it to the bottom-right corner after each use. The update makes it easier to continue conversations without reopening the Gemini app and is now widely available on Android.

RUN CURSOR CLOUD AGENTS ON MACHINES YOU MANAGE

Cursor cloud agents can now run on self-hosted machines inside a team's own network, with a worker linking that infrastructure to Cursor's agent loop. Worker pools scale with queued demand, hibernate idle machines, and run across partners such as AWS Lambda, Cloudflare, Modal, and Vercel. Cloud agents already produce more than 60% of the pull requests Cursor merges internally, and Linux workers now support computer use alongside Macs.

PERCEPTUAL ISSUES IN DYNAMIC INTERFACES

Dynamic interfaces built with Ajax can change content without users ever noticing, because inattentional blindness, change blindness, and attentional gambling let even large updates slip past. In testing on a UK government site, users caught a content update but missed the menu expansion happening at the same time. Making changes bigger, brighter or animated, and placing feedback where users are already looking, keeps updates from going unseen.

THE CRITICAL DESIGN ENGINEERING MANIFESTO

A remix of the 2011 Critical Engineering Manifesto adapts its statements for Critical Design Engineers navigating agentic software tools. Eleven numbered statements set guardrails such as treating generated code as both capability and threat, designing seams instead of seamlessness, and recording the "why" behind decisions. The text lives as a versioned GitHub Gist under the GNU Free Documentation License, with an influence register crediting the thinkers behind each statement.

AI IN PRODUCT TEAMS: IN 2026, THE GROWING IMPACT ON COLLABORATION

AI adoption in product teams is shifting from ad hoc experimentation to more structured, organization-wide strategies, although progress varies widely across companies. Research and examples from firms like Fin, Amplitude, and N26 suggest that clear governance, experimentation, and cross-functional collaboration deliver stronger productivity and teamwork benefits than simply encouraging AI use. At the same time, AI is blurring the boundaries between design, engineering, and product roles, creating both new opportunities and new challenges as companies work out how to adapt their culture, workflows, and responsibilities.

TSPARTICLES: POWERFUL PARTICLE ENGINE, SIMPLE DEVELOPER WORKFLOW

Ship interactive backgrounds, effects, and confetti on any framework. Start with a minimal setup, then scale to presets, plugins, and custom shapes.

SELF-HOSTED GOOGLE FONTS CONFIGURATOR

Your site might be violating GDPR by loading fonts from Google's servers. Self-host them in minutes to keep performance fast and user data private.

AI-POWERED ARCHITECTURE DOCUMENTATION

Archyl maps every system, service, and dependency into an interactive C4 model, then keeps it in sync on every push.

WHAT IF HUMANS ARE NO LONGER AT THE CENTER OF DESIGN?

Planetary Design widens human-centered design to the social, technological, and ecological systems around users, with Kate Raworth's Doughnut Economics as its starting frame. TetraPOT, a wave breaker that shelters mangrove seedlings, and IDEO's pre-, mid-, and post-flight rituals show how artifacts can surface non-human stakeholders and trade-offs. A new University of Michigan course on Design for Longevity applies the lens at the Dexter Senior Center, mapping the ecosystem around older adults.