Search the archive

Schedule a demo to learn more

AMAZON TO ANNOUNCE LARGEST LAYOFFS IN COMPANY HISTORY, SOURCE SAYS

Amazon is expected to start informing employees about sweeping job cuts starting today. It plans to lay off as many as 30,000 staff across its corporate workforce. The layoffs are part of a broader cost-cutting campaign that started during the COVID-19 pandemic. Amazon CEO Andy Jassy said earlier this year that the company's workforce could shrink further as a result of generative AI.

OPENAI'S LESS-FLASHY RIVAL MIGHT HAVE A BETTER BUSINESS MODEL

Anthropic gets about 80% of its revenue from corporate customers, of which it has around 300,000. Its Claude language models have been praised for their aptitude in coding. Anthropic is surprisingly close to OpenAI in revenue. It expects to get to a $9 billion annual run rate by the end of the year, a big lead in revenue per user over OpenAI.

SAM ALTMAN REPORTEDLY DEVELOPING NONINVASIVE BRAIN INTERFACE TO RIVAL ELON MUSK'S NEURALINK

Sam Altman's Merge Labs is a brain-computer interface (BCI) startup. It recently hired an award-winning biomolecular engineer, Mikhail Shapiro. Shapiro's past work suggests that the startup could be aiming to create a device that could link to the brain using sound waves. If Merge Labs is successful, its device may be preferable to Neuralink's BCI as it wouldn't require surgery.

INSIDE THE MINDS BUILDING THE FUTURE: NOTES FROM THE PROGRESS CONFERENCE 2025

The Progress Conference is a four-day event that connects people and ideas in the progress movement. The speakers at the conference discussed a technologically advanced future and provided thoughtful perspective on how we can make progress happen. This post contains notes and thoughts on some of the presentations at the conference, which had over 300 attendees.

EASY RISC-V

This is an introductory tutorial to RISC-V assembly programming intended for those with a basic familiarity with low-level computer science concepts. RISC (Reduced Instruction Set Computer) architecture started its life at UC Berkeley. It is an open standard with a clean design that is quickly gaining steam and has found great success in many areas of application. The tutorial covers the 32-bit bare bones RV32I_Zicsr instruction set with a tiny subset of the privileged architecture.

WHY IP ADDRESS TRUNCATION FAILS AT ANONYMIZATION

Truncated IP addresses are still personal data under GDPR, especially when combined with other information that most systems routinely collect. Subnet size doesn't correlate with user anonymity. The network prefix itself contains enough information for trivial identification. Many organizations are creating compliance risks by thinking IP truncation is GDPR-compliant anonymization. What is actually needed is encryption specifically designed for IP addresses.

MORE BIG COMPANIES BET THEY CAN STILL GROW WITHOUT HIRING

Companies in the US are increasingly making the calculation that they can keep the size of their teams flat, or even shrink them, without harming their businesses. Part of this is the belief that artificial intelligence will be used to automate more processes, but companies are also hesitant to make moves in an uncertain economy. The widespread caution in hiring is frustrating job seekers and leading many employees within organizations to feel stuck in place. Operating with fewer people can pose risks for companies by straining existing staff or hurting efforts to develop future leaders.

MUSK'S WIKIPEDIA RIVAL SITE LIVE AFTER CRASHING ON LAUNCH DAY

Grokipedia crashed shortly after it launched yesterday. It was live late evening with nearly 900,000 articles. Many entries closely resemble their Wikipedia counterparts, but Elon Musk's views shape how some topics are framed. For example, his own Grokipedia page features a 'Recognition and Long-Term Vision' section instead of Wikipedia's 'Accolades'.

Subscribe now.

SAUDI ARABIA, RICH WITH OIL, WANTS TO BE KNOWN AS THE AI EXPORTER

The kingdom is building three major data center complexes aimed at foreign companies.

QUALCOMM LAUNCHES AI CHIPS TO CHALLENGE NVIDIA'S DOMINANCE

Qualcomm's AI200 will start shipping next year, and the AI250 in 2027.

THE RENAISSANCE OF WRITTEN CODING CONVENTIONS: BECAUSE AI READS MANUALS, TOO

AI learns from whatever code bases you give it, and if your codebase is inconsistent or chaotic, the AI will reflect that.

TOWARDS A CONTEXT SHARING RUNTIME FOR THE PERSONALISED WEB

Instead of tightly controlling which applications have access to data, we should control where applications can send it.

JSON QUERY

JSON Query is a lightweight, flexible, and expandable JSON query language.

META BURIES THE METAVERSE: ZUCKERBERG TAPS VISHAL SHAH TO LEAD HIGH-STAKES AI REVOLUTION

Vishal Shah, a longtime executive who once led Meta's metaverse experiment, has joined Meta Superintelligence Labs to bridge the gap between cutting-edge AI research and Meta's apps.

Get the report ↗️

ADVANCING CLAUDE FOR FINANCIAL SERVICES

Anthropic is continuing its push into financial services by launching Claude for Excel, which lets users analyze and modify spreadsheets by chatting with Claude in a sidebar. It is currently in beta behind a waitlist.

THE AI STARTUP FUELING CHATGPT'S EXPERTISE IS NOW VALUED AT $10 BILLION

Mercor, a human resources and recruiting startup, has finalized a new funding deal that values the company at $10 billion. The company manages 30,000 contractors around the world who help AI chatbots learn how to think and speak like humans. Venture capital firms are eager to offer funding to founders at sky-high valuations because they fear missing out on the next AI tech giant. Mercor is currently being sued by Scale for allegedly stealing trade secrets.

SPEEDRUNNING AN RL ENVIRONMENT

RL environments can be surprisingly complex and fun to create. This post explains what RL environments are, introduces the 'verifiers' framework, and also walks readers through how to create an environment for a benchmark called AgentDojo. RL environments are scenarios that LLMs operate in for evaluation or training. Designing them means essentially defining the maze, the rewards, and how the LLM navigates through it.

AI DISCOVERS NOVEL CANCER DRUG, OR DID IT?

Google recently published an announcement that its Gemma model had discovered a new potential cancer pathway. It's important to note that the model didn't invent or theorize a method to discover potential drug candidates - the entire construction of the experiment was imagined and executed by humans. The model was used to probabilistically narrow a set of potential candidates to a manageable set that could be inspected by human review. Humans still possess a reasoning capability that is not present in any current AI system.

TOKEN-ORIENTED OBJECT NOTATION (TOON)

TOON is a compact, human-readable format designed for LLM input that allows developers to pass structured data to LLMs with significantly reduced token usage. It excels at uniform complex objects. Standard JSON is verbose and token-expensive - TOON conveys the same information with fewer tokens. In benchmark tests, TOON achieves higher accuracy compared to JSON while using significantly fewer tokens.

ON-POLICY DISTILLATION

Thinking Machines Lab demonstrated that training smaller AI models by having them learn from their own mistakes, by grading them with a larger teacher model, achieves the same reasoning performance as RL at 9-30x lower cost. "On-policy distillation" combines the relevance of learning from your own outputs with the dense feedback of traditional distillation, reaching 70% on AIME'24 math problems in just 150 training steps compared to 17,920 GPU hours of RL.

REASONING WITH SAMPLING

A new sampling method enables base models to reach single-shot reasoning performance comparable to reinforcement learning, without retraining or external verifiers, while maintaining diversity and multi-shot accuracy.

QUALCOMM ANNOUNCES NEW AI CHIPS IN DATA CENTER PUSH, SHARES SURGE

The smartphone chipmaker's new AI200 and AI250 inference accelerators—shipping in 2026 and 2027—sent shares up 20% on the promise of diversification beyond a stagnating mobile market. Saudi Arabia's Humain is committed to purchasing roughly $2 billion worth of the chips starting next year.

MINIMAX M2

Chinese Lab MiniMax open-sourced M2, a 230B parameter model with 10B active that ranks #1 among open-source models on Artificial Analysis' composite intelligence score. It slightly outperforms the previous generation Claude Sonnet model on coding benchmarks at less than 10% of the cost.

started in seconds.

CHATGPT'S MOBILE APP IS SEEING SLOWING DOWNLOAD GROWTH AND DAILY USE, ANALYSIS SHOWS

ChatGPT's mobile app download growth and daily active use are slowing.

AI-POWERED SEARCH ENGINES RELY ON “LESS POPULAR” SOURCES, RESEARCHERS FIND

AI search engines tend to cite less popular websites - ones that wouldn't even appear in the Top 100 links listed in an 'organic' Google search.

STRENGTHENING CHATGPT'S RESPONSES IN SENSITIVE CONVERSATIONS

Following recent safety controversies, OpenAI revealed that approximately 1-3 million users per week show signs of mental health emergencies, suicidal ideation, or emotional reliance on ChatGPT.

RATE-CUT OPTIMISM FUELS $921M IN GLOBAL CRYPTO FUND INFLOWS

Crypto investment products saw $921 million in weekly inflows, rebounding sharply from the prior week's $513 million outflow. The increase was driven by rate-cut optimism after lower CPI data and strong US demand, with Bitcoin funds adding $931 million, their largest gain since rate cuts began, while Ethereum products saw $169 million in outflows, their first dip in five weeks. Trading volumes held strong at $39 billion, well above average, and inflows were led by the US ($843M) and Germany ($502M).

CITI, COINBASE PARTNER TO EXPAND DIGITAL ASSET PAYMENTS

Citi partnered with Coinbase to develop digital asset payment solutions for institutional clients, initially focusing on streamlining fiat on/off-ramps and payment orchestration with future plans for 24/7 fiat-to-stablecoin options. The partnership extends Citi's push into round-the-clock digital money solutions alongside existing offerings like Citi Token Services and USD Clearing across its 94-market network. Citi currently banks 90% of top e-commerce firms and 15 of the world's 20 largest fintechs. It plans to launch crypto custody services in 2026 to hold native digital assets like Bitcoin and Ethereum on behalf of clients.

JAPAN'S JPYC LAUNCHES FIRST LEGALLY APPROVED YEN STABLECOIN

JPYC Inc. has launched Japan's first legally recognized yen-backed stablecoin, fully compliant under the country's Payment Services Act. Pegged 1:1 to the yen and backed by bank deposits and government bonds, the JPYC token is live on Avalanche, Ethereum, and Polygon, with issuance and redemption handled via the new JPYC EX platform. Users can acquire tokens after KYC through Japan's My Number ID system. JPYC targets ¥10 trillion ($65B) in circulation within three years. Multiple Japanese firms plan to integrate JPYC into retail and enterprise payment systems, signaling Japan's entry into regulated, large-scale stablecoin adoption.

STABLECOIN LINKED CARDS - WAY BETTER SECURED CREDIT CARDS

Stablecoin-linked cards allow users to spend USDC or USDT at any Visa/Mastercard merchant through instant fiat conversion at point of sale. Self-custodial options like Ether.fi allow users to borrow against crypto collateral at 4% APY while earning 10% yield. Revolut's zero-fee stablecoin conversion for 65 million customers and Coinbase's Bitcoin-backed loans powered by onchain protocols like Morpho demonstrate how vaults are bringing decentralized lending to mainstream fintech apps, replacing balance sheet lending with private credit models. The shift toward direct stablecoin acceptance by merchants and banks will compress settlement times to zero, eliminating the 1-2% early payment fees merchants currently pay and the up to 2% consumer on/off ramp costs.

WHY X402 MATTERS

x402 transforms how AI agents pay for data and services by replacing traditional payment rails, which are slow, risky, and require human setup, with instant, trustless crypto settlements. Using USDC microtransactions, agents can autonomously authorize and complete payments directly with APIs, removing middlemen and chargebacks. This shift enables fully autonomous AI agents and lets API providers design new onchain business models where data, compute, and services can transact seamlessly in real time.

A NEW PATH FOR PROPRIETARY AMMS

Proprietary AMMs managed by professional market makers have captured 40%+ of Solana trading volume by updating pricing curves up to 74 times per second, enabled by extremely cheap updates compared to swaps that allow market makers to achieve Top of Block priority with minimal fees and avoid front-running. EVM's expensive storage writes versus swaps create only a 1:10 ratio instead of Solana's 1:1000, making updates costly and vulnerable to arbitrageur front-running since outbidding requires only 10x the priority fee. A proposed solution combines a Global Storage smart contract for price curve parameters with a builder policy that reserves the first 5-10% of block gas exclusively for these updates ordered by priority fee, creating a protected "fast lane" that guarantees Top of Block execution and establishes a local fee market competing only against other updates rather than all DeFi transactions.

LOCKONOMICS

Velodrome and Aerodrome – the leading DEXes on Optimism and Base – pioneered "lockonomics," a model where users lock tokens for up to 4 years to receive voting power, protocol fees, and project bribes, while LPs receive token emissions instead of direct fees. The system imposes a "speculation tax" on short-term holders through inflation that benefits lockers – currently 50% of supply is locked with an average duration of 3.6 years – and can be stabilized via emission control to match revenue when growth slows. Velodrome currently runs 15% annualized inflation while delivering 20%+ APY to lockers, deliberately operating at emissions exceeding revenue to drive sticky growth through accumulated protocol-owned liquidity and user lock-in before eventually reverting to a sustainable emissions-equal-revenue equilibrium.

WTF IS ORCHESTRATION (OR CLEARING) (OR SETTLEMENT)

Orchestration (coordinating between parties), clearing (reconciling who owes what), and settlement (actual fund transfers) evolved in traditional finance to coordinate transactions across networks of intermediaries maintaining separate ledgers. Crypto's multi-chain reality has recreated the same coordination challenges that shaped traditional finance, despite blockchain's promise of instant settlement on a single shared ledger. The opportunity lies in rebuilding these functions through smart contracts, cryptographic proofs, and programmable liquidity rather than institutional intermediaries, creating systems that are more performant, upgradeable, and transparent to handle crypto's expected orders-of-magnitude higher velocity of money.

POLYGON'S B2B PAYMENT VOLUME MOVES TOWARD $500M

B2B payment projects built on Polygon have seen rapid adoption, with monthly volume rising from $140M in January to nearly $300M in August, rebounding after a brief dip in September.

KAST SCALES TOWARD BECOMING THE “GLOBAL CRYPTO/FIAT NEOBANK”

KAST has built 580 product versions in just 484 days.

INDIAN COURT BARS WAZIRX FROM USING CUSTOMER XRP TO OFFSET HACK LOSSES

The Madras High Court ruled that cryptocurrency is property.

MEGAETH RAISES $50M IN MINUTES AS MEGA SALE SELLS OUT 3X OVER

MegaETH, the Ethereum Layer 2 project, sold out its $50 million MEGA token sale within minutes, ending a planned 72-hour auction early after demand exceeded supply by over 3x.

RESTYLE YOUR INSTAGRAM STORIES USING META AI TO TAKE YOUR CREATIVITY FURTHER

Instagram has introduced a new restyle feature, powered by Meta AI, that lets users edit photos and videos in Stories by adding, removing, or changing elements via text prompts or preset effects. Users can make simple edits like removing backgrounds or adding accessories, apply creative effects like watercolor or film noir, and even transform video backgrounds into underwater or snowy landscapes. The feature includes Add Yours stickers that enable friends to replicate edits and start viral trends across the platform.

NETFLIX GOES ‘ALL IN' ON GENERATIVE AI AS ENTERTAINMENT INDUSTRY REMAINS DIVIDED

Netflix is embracing generative AI as a production tool. CEO Ted Sarandos maintains it won't replace creative storytelling, citing recent uses in shows like "The Eternaut" and "Happy Gilmore 2" for visual effects and de-aging. The streaming giant's AI adoption comes amid industry-wide concerns about job displacement and unauthorized use of artists' work in training data, with recent controversies involving OpenAI's Sora 2 model generating deepfakes of actors. Despite positioning itself as "all in" on AI technology, Netflix reported quarterly revenue of $11.5 billion, falling short of its forecast despite 17% year-over-year growth.