Search the archive

WORLD MODELS: COMPUTING THE UNCOMPUTABLE

This is a comprehensive guide that covers the history, theory, progress, and potential of World models.

WE NEED TO TALK ABOUT AGENTS

When everything becomes an 'agent,' it's nearly impossible for buyers to tell what any given product actually does, how it fits into existing operations, how they should interact with it, or why it matters to their bottom line.

BE INTENTIONAL ABOUT HOW AI CHANGES YOUR CODEBASE

Using the right patterns will lead to scalable codebases, which leads to faster iteration cycles and better software.

FULLY REMOTE)

As the first engineer on TLDR's new Applied AI team, you'll build AI agents to automate end-to-end workflows and lead our context engineering project to allow non-technical teammates create their own AI workflows. Learn more [21].

XIAOMI LAUNCHES NEXT-GEN SU7 WITH 902 KM RANGE AND LIDAR, STILL UNDERCUTS TESLA

Xiaomi's SU7 electric sedan, which starts at around $33,000, has up to 902 km of CLTC range, standard LiDAR across all trims, and an 897V charging architecture on the top model.

THOUGHTS ON OPENAI ACQUIRING ASTRAL AND UV/RUFF/TY

In the worst-case scenario, the projects can be forked, and the community can take over maintenance.

Get started with Enterprise Search

CURSOR COMPOSER 2

Cursor has released Composer 2, a coding model offering frontier-level performance at significantly reduced token pricing.

OPENAI TO ACQUIRE ASTRAL

OpenAI announced it will acquire Astral to integrate popular Python tooling into Codex and expand AI-assisted software development capabilities.

PERPLEXITY LAUNCHES PERPLEXITY HEALTH AGENT IN US

SCALING KARPATHY'S AUTORESEARCH: WHAT HAPPENS WHEN THE AGENT GETS A GPU CLUSTER

Researchers pointed Claude Code at autoresearch and gave it access to 16 GPUs on a Kubernetes cluster. Over 8 hours, it submitted around 910 experiments. The parallelism changed how the agent searched - with one GPU, it was stuck doing greedy hill-climbing, but with 16 GPUs, it ran factorial grids of 10 to 13 experiments per wave, catching interaction effects between parameters that sequential search would miss.

MONITORING AUTONOMOUS CODING AGENTS

OpenAI described a monitoring system for internal coding agents designed to detect misalignment risks and study real-world agent behavior in tool-rich environments.

OPENCLAW WAS THE WORDPRESS MOMENT

OpenClaw showed the promise of autonomous AI agents but failed in production due to context management, non-determinism, weak tooling, and model limitations. Fully agentic systems are fragile, with real-world use cases relying on structured workflows plus LLM steps rather than pure autonomy. Next wave shifts to vertical “agent harnesses” that package domain-specific context, reliability, and infra into usable products.

WORLD MODELS: COMPUTING THE UNCOMPUTABLE

World Models are advancing AI by simulating real-world complexities through action-conditioned neural networks, enabling efficient prediction and planning. Generative and latent approaches to World Models are driving breakthroughs across various applications, from robotics to autonomous driving, leveraging vast datasets like gaming clips to approximate human decision-making. Major investments from industry leaders, including companies like General Intuition and World Labs, underscore the significant potential and rapid evolution of World Models in achieving general intelligence.

FREE YEAR → framer.link/TLDRAI

Startups move fast, and design should too. With Framer,

early-stage companies can launch a beautiful, production-ready site in hours. Now you can get one year of Framer Pro free [14] and join thousands of founders already building on Framer. APPLY TO CLAIM YOUR

AGENT AUTH PROTOCOL

The Agent Auth protocol makes the run-time agent a first-class principal. Each agent is registered with its own identity, granted specific capabilities, and governed by a lifecycle that the server controls. The server sees which agents are acting, what they are authorized to do, and can terminate agents without affecting anything else. The protocol is designed for existing infrastructure and is easy to adopt.

NANOGPT SLOWRUN: 10X DATA EFFICIENCY WITH INFINITE COMPUTE

Researchers achieved 10x data efficiency with NanoGPT Slowrun, a benchmark for language modeling algorithms in the infinite compute, within a few weeks. Data efficiency matters because compute grows much faster than data. Intelligence will eventually be bottlenecked by data, not compute. This data efficiency result allows researchers to improve model performance by scaling with compute rather than with data.

MOLMOPOINT FROM AI2

MolmoPoint is a new open-source grounding architecture with three models, a 36K-image GUI dataset, and new tracking data for video and interface tasks.

GOOGLE DEEPMIND'S ONLINE RLHF WITH 10X DATA EFFICIENCY

An online RLHF algorithm incrementally updated reward and language models, achieving over 10x data efficiency gains using uncertainty modeling and information-directed exploration.

BROAD TIMELINES

AI timelines remain uncertain, with experts divided on when AI will significantly change the world. The best approach isn't to pick a specific year, but to work with broad timelines that account for diverse expert opinions. Decision-makers should hedge against both short and long-term scenarios, investing in strategies adaptable to rapid changes or gradual shifts in AI development.

RETHINKING OPEN SOURCE MENTORSHIP IN THE AI ERA

AI contributions have increased noise in open source, making traditional mentorship signals unreliable. GitHub has introduced the '3 Cs' framework (Comprehension, Context, and Continuity) to help maintainers identify worthy contributors for mentorship. This strategy aids in maintaining mentorship efficacy and ensuring the growth and sustainability of open source communities amidst AI advancements.

JENSEN HUANG DOESN'T NEED A NEW CHIP. HE NEEDS A NEW MOAT

Nvidia's CEO Jensen Huang is aiming to transform the company from a chipmaker to a dominant AI platform operator through NemoClaw, an open-source platform for AI agents. This move minimizes reliance on chip sales cycles and positions Nvidia for more enduring, platform-based growth. However, success hinges on enterprise adoption and outpacing rapid advancements from Chinese competitors.

THE CURRENT AND PROJECTED IMPACT OF AI AND FORMALIZATION ON THE PRACTICE OF MATHEMATICS IS ANALOGOUS TO THE IMPACT THAT THE AUTOMOBILE HAD ON THE EVOLUTION OF CITIES

Thoughtful urban planning, as well as the development of social and legal rules on how to manage traffic, were necessary to allow pedestrian and automotive transport to co-exist in a way that retains the benefits of both.

WE ANALYZED HUNDREDS OF FORMER OPENAI EMPLOYEES TO SEE WHERE THEY GO NEXT

Besides the big-name AI labs, ex-OpenAI employees are going to an under-the-radar company called Periodic Labs.

THE NEED FOR AN INDEPENDENT AI GRID

It is difficult to scale compute without compromising independence at scale.

CHARACTER.AI'S IMAGE GALLERY

Imagine Gallery is a feature that organizes and manages generated visuals with filtering, sharing, and customization options within the app.

GOOGLE IS REPORTEDLY TESTING A GEMINI APP FOR MAC

The macOS app will offer the same set of features as Gemini on the web, but it will also have Desktop Intelligence, giving it a new source of information and context for responses.

TEMPO MAINNET GOES LIVE WITH MACHINE PAYMENTS PROTOCOL

Tempo Mainnet launched on March 18, providing high-throughput, instant-settlement payment infrastructure, paired with the Machine Payments Protocol (MPP), co-authored by Stripe and Tempo as the protocol layer for programmatic agent-to-service payments. MPP defines a request-response flow in which an agent requests a resource, receives a payment challenge, authorizes payment, and receives delivery, with stablecoin support, card rails via Visa and Stripe, and Bitcoin Lightning via Lightspark, plus a Sessions primitive that lets agents pre-authorize recurring micropayments within defined spending limits. The protocol launched with a directory of 100+ services and institutional partners, including Anthropic, OpenAI, Visa, Mastercard, Shopify, and Revolut. Stripe users can begin accepting MPP payments through the existing PaymentIntents API, with MPP support sitting alongside x402 in Stripe's broader Agentic Commerce Suite. Stripe has submitted MPP to the IETF as an open internet standard, a move that positions the protocol as a candidate for standardized machine payment infrastructure across the broader web.

S&P 500 PERPETUAL CONTRACT LAUNCHES ON HYPERLIQUID

S&P Dow Jones Indices and trade[XYZ] have launched the first official S&P 500 perpetual contract, listed on Hyperliquid and accessible 24/7/365. SPDJI's provision of official index data separates this product from synthetic S&P 500 instruments, giving institutional participants the benchmark integrity required to support deep liquidity at scale. The contract removes exchange operating hours, geographic restrictions, and intermediary dependencies that have long constrained exposure to the benchmark.

X402 ADDS UNIVERSAL ERC-20 TOKEN SUPPORT

The x402 payment protocol has extended beyond its original EIP-3009 foundation to support virtually any ERC-20 token, integrating Uniswap's Permit2 alongside two new gas sponsorship extensions that handle transaction fees in the background. EIP-2612 Gas Sponsorship covers tokens with native permit functionality, while ERC-20 Approval Gas Sponsorship handles standard tokens such as USDT, allowing facilitators to abstract gas costs from end users across both paths. The CDP facilitator already supports these extensions on Base and Polygon, making the upgrade immediately usable rather than theoretical.

ERC-8183: EVALUATED AGENT COMMERCE

ERC-8183 proposes a three-party job lifecycle for onchain agent commerce where a client escrows funds, a provider submits work, and an independent evaluator determines settlement, making payment conditional on verified output rather than execution alone. The standard complements ERC-8004 by handling current-task verification at the moment of settlement, since reputation systems operate on historical outcomes and cannot assess whether a specific task meets its defined criteria. Evaluated outcomes feed back into reputation signals over time, improving agent routing and trust, with non-deterministic tasks such as research quality, trading execution, and data strategy requiring domain-specific evaluator agents to apply task-appropriate rubrics.

RWA-BACKED LENDING: THE REAL DEFI 2.0

Crypto-backed DeFi lending has revealed a structural ceiling, with USDC borrow rates on Aave collapsing below 2% in early 2026 (under the 3.7% Fed funds rate), confirming that crypto-collateral demand is tied to leverage cycles rather than broader credit markets. RWA collateral on Morpho commands a 50-80% rate premium over crypto collateral (4.4-5.7% vs. 3.2-3.4%), with $750M in RWA collateral supporting $330M in stablecoin borrows, a spread driven by looping demand from yield-seekers arbitraging real-world asset returns on-chain. 3F, a one-click RWA leverage product on Morpho, automates these borrow loops using collateral from BlackRock, Fidelity, Franklin Templeton, and Baillie Gifford, marking a shift toward institutional-grade collateral in on-chain credit markets. RWA-backed yields are anchored to global credit markets rather than crypto sentiment, so borrow rates in this segment hold through bear market cycles, addressing the core volatility problem that crypto-only DeFi has yet to solve.

MAPPING OUT CRYPTO'S MIDTERMS FATE

Bipartisan floor support for crypto legislation, including 48% of House Democrats and 40% of Senate Democrats backing the GENIUS Act, obscures a more hostile committee dynamic that would govern under a Democratic majority. Maxine Waters, probable House Financial Services chair if Democrats flip the chamber (analysts put that probability at 85%), has voted against every major crypto bill and controls which legislation receives hearings or markups, requiring no floor vote to block bills outright. Elizabeth Warren's Senate Banking chairmanship remains the binding constraint at the upper chamber, though Ruben Gallego's expected leadership of the Digital Assets subcommittee positions the Senate as the more viable legislative venue for industry priorities. Fairshake's $7M campaign against Juliana Stratton in the Illinois Democratic primary failed, suggesting crypto PACs face meaningful limits in reshaping the primary electorate that determines committee composition.

CRYPTO/STABLECOIN INFRASTRUCTURE: HIDDEN GIANTS MOVING BILLIONS

BVNK processed $16.2 billion in crypto/stablecoin transfer volume in 2025, exceeding Revolut's $14.3 billion despite holding only a 6/10 brand awareness score compared to Revolut's 8.5/10. Visa's crypto card and stablecoin partners including Rain ($1.24B), RedotPay, Reap, and ether.fi Cash combined for $4.9 billion in volume. In crypto infrastructure, volume and visibility often diverge: Blindpay moved $500 million with almost zero public profile, underscoring that the most important payment rails are frequently invisible.

CRYPTO HACKS AVERAGE $25 MILLION AS LARGEST EXPLOITS SKEW LOSSES

Immunefi data shows the average crypto hack nets around $25 million, but the industry's headline loss figures are heavily skewed by a handful of massive exploits. Nearly 70% of 2025's approximately $17 billion in total losses came from just three incidents, with one breach alone exceeding $1.4 billion. Over 90% of projects still harbor critical, exploitable vulnerabilities despite Immunefi's 45,000-researcher bug bounty community having saved an estimated $25 billion in user funds across protocols including Polygon, Optimism, Chainlink, and Synthetix.

SEC OFFICIALLY ADOPTS "ONCHAIN" OVER "ON-CHAIN"

An SEC guidance published on March 17 uses "onchain" as a single word in its footnotes, settling the stylistic debate in favor of the unhyphenated form.

CHAINLINK INTEGRATES EUROPE'S LARGEST ASSET MANAGER

Europe's largest asset manager and its tokenized fund platform have integrated Chainlink infrastructure.

ETHEREUM HOLDS 54% OF TOTAL STABLECOIN SUPPLY

Ethereum commands 54% of total stablecoin supply, followed by Tron at 28%, Solana at 5.4%, and BNB Chain at 5.1%.

DELPHI DIGITAL: STABLECOINS THREATEN BANKING PROFIT MODEL

Treasuries yield roughly 3.89% while standard savings accounts pay 0.39%.

SPOTIFY'S WEAR OS OVERHAUL BRINGS CLEANER VISUALS AND QUICKER CONTROLS TO YOUR WRIST

Spotify has launched a major visual overhaul of its Wear OS app featuring cleaner visuals, swipe-based navigation, and easier access to playback controls. The redesign includes a music-first mode that highlights album art, simplified tap-and-double-tap controls, and quick access to key sections like Home, Search, and Your Library. The update is rolling out via the Google Play Store over the coming days, with more features planned for future releases.

DESIGN UI USING AI WITH STITCH FROM GOOGLE LABS

Google Labs has evolved Stitch into an AI-native design canvas that lets users create high-fidelity UI designs from natural-language descriptions via "vibe design." The platform features an infinite canvas, a design agent that reasons across entire projects, and voice capabilities for real-time collaboration and iteration. Stitch enables rapid prototyping by transforming static designs into interactive prototypes instantly and integrates with other workflow tools through its MCP server and SDK.

YOU CAN NOW TEACH ADOBE FIREFLY TO GENERATE IMAGES IN YOUR OWN VISUAL STYLE

Adobe has launched Firefly custom models in public beta, allowing users to train AI on their own visual style using 10 to 30 images, enabling consistent generation of images that match their aesthetic across projects. The feature is available to premium subscribers and costs 500 credits per training, with support for photography, illustration, and character styles.

DESIGN PROCESS ISN'T DEAD, IT'S COMPRESSED

The recent argument to abandon the design process in favor of intuition and rapid building misrepresents how experienced designers actually work. What appears to be "skipping process" is actually process compression - experienced designers internalize and adapt traditional frameworks, moving fluidly through discovery, ideation, and evaluation stages. Solution-first approaches only work effectively in mature problem spaces with established patterns, not when inventing something entirely new.

GENERATIVE UI: USEFUL, OR JUST CHAOS WITH BETTER BRANDING?

Generative UI refers to interfaces dynamically created by AI in response to user input or context, rather than using fixed screens and components. While it offers benefits like personalization, reduced friction, and context-awareness, it risks undermining user confidence through unpredictability and breaking muscle memory. Success requires strict adherence to guardrails, including design system constraints, user control options, accessibility planning, and stable core navigation patterns.

DESIGNERS, THERE'S NO EXCUSE FOR ACCESSIBILITY AS AN AFTERTHOUGHT

Accessible design is often overlooked in graphic design due to habits, assumptions, and late-stage implementation, but it should be a core part of the process because designing inclusively leads to better outcomes for everyone. Start accessibility early, consult people with lived experience, and use simple practices—like not relying on color alone—to make designs more effective and inclusive.