Search the archive

Curious? Try Wispr Flow Free

META PLATFORMS ACQUIRES HUMANOID ROBOT STARTUP ASSURED ROBOT INTELLIGENCE

Meta has acquired a startup that is trying to build humanoid robots. Assured Robot Intelligence is a company at the frontier of robotic intelligence. Its team will join Meta to focus on optimizing models for robotics applications. Investors are skeptical of Meta's spending plans - the company's stock was down following the announcement of the acquisition.

SPACEX HAS SPENT MORE THAN $15 BILLION ON STARSHIP AND IS RACING TO MAKE ROCKETRY RESEMBLE AN AIRLINE SCHEDULE

SpaceX has so far spent more than $15 billion developing Starship. The company's capital expenditure surged nearly fivefold between 2024 and 2025 to $20.7 billion. Of that spend, $12.7 billion was directed at AI initiatives. Starlink is still the company's financial engine, making $11.4 billion in revenue in 2025.

THE COMPANY THAT BUILT TIKTOK'S ALGORITHM IS NOW DESIGNING DRUGS FOR DISEASES PHARMA CALLED UNDRUGGABLE

ByteDance has entered the AI drug discovery race. The company's drug discovery unit, Anew Labs, recently presented its first AI-designed therapy at a major immunology conference. Its work targeted a protein-protein interaction long thought to be undruggable. The unit has published a generative model that it claims can design functional molecules across all scales.

AMERICA'S ELECTRICITY GAP

Power access is incredibly important to tech companies' gamble on AI development. The US was always going to need more power investment to meet its next phase of economic growth and was already struggling to meet rising electricity demand well before the AI boom. The costs of the electricity gap are mounting as power becomes more important to the national economy. More of the US will have to permit the fast deployment of solar and batteries if they want to meet their own increasing electricity needs.

Vercel, and Clay use Flow daily. Try it free

AGENT SKILLS

AI coding agents are extremely capable, but they have no instinct for the parts of the job that don't show up in the diff. Agents will skip the senior-engineering work unless you make it impossible to skip. The job increasingly is to encode discipline as something agents can't talk themselves out of. Skills are one way to do that. The senior-engineer parts of the job are no longer optional, even when the engineer is a model.

OPEN DESIGN

Open Design is an open-source alternative to Claude Design. It uses the same loop and artifact-first mental model with none of the lock-in. Open Design is local-first, web deployable, and supports several coding agents. It behaves like a senior designer with a working filesystem, a deterministic palette library, and a checklist culture.

SOMETIMES SMART PLANNERS LOSE TO SIMPLE MARKETS

OPENAI WANTS TO GO PUBLIC. FIRST SARAH FRIAR NEEDS TO GET IT TO GROW UP

Sarah Friar was brought into OpenAI to give it the financial maturity needed to go public. Over the past few months, she has balanced the company's spending expectations and questioned the wisdom of spending even more money on data centers. It is up to Friar to pull off what could be one of the biggest IPOs ever. If she moves too slowly, OpenAI could fall behind and hit the market after Anthropic. Moving quickly risks stretching the business too far.

89% of messages with zero edits. Try free

THE DUALITY OF LANGUAGE MODELS IN THE BROWSER

In-browser models are free and private, but they only work well if you can afford a decent device with a GPU.

HOW MIGHT A BROWSER BE DEVELOPED?

Browsers and web development will change as large language models evolve.

PROMPT ENGINEERING IS PERMANENT

Prompt engineering today is about making AI systems reliable when they have context, tools, memory, permissions, and real consequences.

BECOME A CURATOR FOR TLDR AI (3-5 HRS/WEEK)

TLDR is looking for an engineer/researcher at a major AI lab or startup to help write for 1M+ subscribers. Our curators have been invited to Google I/O and OpenAI DevDay, scouted for Tier 1 VCs, and get early access to unreleased TLDR products. Learn more [21].

SPECSMAXXING

Spec-writing is where the act of software engineering really happens.

GEMINI APP GETTING FULL REDESIGN THAT OVERHAULS EVERY PART OF THE UI

Screenshots of the new design are available in the article.

SCROLLING IS ON THE DECLINE

People of all ages have cut back on the average number of hours spent scrolling on social media per day.

Learn more about evals →

ANTHROPIC TESTS JUPITER-V1-P AHEAD OF ITS DEVELOPER CONFERENCE

Anthropic appears to have started a fresh round of red teaming on a new internal build. The company is set to host its Code with Claude developer conference in San Francisco on May 6. The timing suggests that the model is being hardened ahead of an announcement timed to the event. The red team round is consistent with the company's responsible scaling policy, which calls for jailbreak probes and constitutional classifier stress tests before any frontier-class deployment.

GOOGLE IS TESTING NEW OMNI MODEL FOR VIDEO GENERATION

Google is testing a new Omni model for video generation, potentially unifying its video and image-generation tools. The Omni model appears in Gemini's video generation UI, suggesting it might become a public product name. A launch during Google I/O 2026 is possible amidst increasing AI video competition.

OPENAI ADDS ANIMATED PETS AND CONFIG IMPORTS TO CODEX

OpenAI updated Codex with animated Pets, which appear as overlays on the screen and interact via short message bubbles. Codex also now auto-imports configuration files from other coding agents and features a new dictation dictionary to improve voice input accuracy. These updates aim to enhance Codex's usability and appeal as a comprehensive desktop application.

DEEPSEEK V4—ALMOST ON THE FRONTIER, A FRACTION OF THE PRICE

DeepSeek's latest preview models, DeepSeek-V4-Pro and DeepSeek-V4-Flash, are both one-million-token Mixture of Experts models. Pro has a total of 1.6 trillion parameters, with 49 billion active, while Flash has 284 billion total, with 13 billion active. DeepSeek-V4-Pro is now the largest open weights model. It is also a very cheap model to run.

CODING PLAN COMPARISONS BASED ON ACTUAL USAGE

This post looks at how pricing compares across coding plans and APIs. Codex is heavily subsidized compared to others, but most others are still subsidized. Claude Pro costs around 10 times more per token than the rest.

HOW DID ‘LARGE' LANGUAGE MODELS GET THAT WAY? THE ROLE OF TRANSFORMERS AND PRETRAINING IN GPT

Large language models are set to be one of the largest computing infrastructure projects ever. This post is the first part of a series about LLM architecture and its implications for reasoning. It looks at why the transformer architecture was so impactful for LLMs.

DESIGNING, REFINING, AND MAINTAINING AGENT SKILLS AT PERPLEXITY

Perplexity emphasizes modular Agent Skills for enhancing its frontier agent products, with specific designs and hierarchies to ensure high-quality user experiences. Unlike traditional software, Skill development prioritizes detailed, context-specific design principles where real queries and evaluations shape their necessity and content. Maintaining these Skills involves constant iteration, testing across multiple models, and prioritizing efficiency and simplicity due to the inherent 'cost' each Skill introduces.

guide

AUTOROUND

AutoRound is an advanced quantization toolkit designed for large language models and vision-language models. It achieves high accuracy at ultra-low bit widths with minimal tuning. AutoRound seamlessly works with Transformers, vLLM, SGLang, and more. It can quantize 7B models in 10 minutes on a single GPU.

SYNTHETIC COMPUTER ENVIRONMENTS FOR AGENT TRAINING

A scalable method generates realistic virtual computer environments and long-horizon simulations, producing rich training signals that improve agent performance across productivity tasks.

REASONING-BASED REWARDS FOR IMAGE EDITING

Edit-R1 introduced a chain-of-thought reward model that evaluates image edits through structured reasoning, improving alignment and performance in text-guided editing tasks.

REPLIT'S AMJAD MASAD ON THE CURSOR DEAL, FIGHTING APPLE, AND WHY HE'D RATHER NOT SELL

Replit's Amjad Masad highlights strong growth, nearing a billion-dollar run rate, and boasts a 300% net revenue retention rate. Unlike Cursor, which struggles with negative margins, Replit maintains gross margin positivity and appeals to non-technical users with its secure, end-to-end platform. While Masad remains committed to Replit's independence, he acknowledges open discussions with potential acquirers and expresses frustration with Apple's alleged discriminatory App Store practices, suggesting possible legal action.

HOW LLM INFERENCE WORKS

This post walks through the inference pipeline from tokenization and embeddings through stacked self-attention layers, then splits generation into two distinct phases on the same GPU: compute-bound prefill that processes all input tokens in parallel and memory-bound decode that emits one token at a time.

YOU ARE NOT IMMUNE TO MODE COLLAPSE

Mode collapse occurs when models repeatedly generate the most common outputs, leading to homogenous results, exemplified by AI generating more dogs over cats with unbalanced training data. It similarly impacts various domains like grant-making and music, as systems become increasingly specialized over time based on prior outputs and successes. To counteract, introduce variability or change external pressures to diversify and prevent over-specialization.

free with code: TLDR1MO. See why people love Granola

HUGGING FACE'S CLEM DELANGUE: STOP COMPARING ENGINES TO CARS

Comparing open-source models to closed APIs is flawed, as they serve different purposes.

TOP AI COMPANIES AGREE TO PENTAGON DEALS FOR CLASSIFIED WORK

Many of the companies have said their deals with the Department of War include commitments that their tools wouldn't be used for mass surveillance or autonomous weapons.

BECOME A CURATOR FOR TLDR AI (3-5 HRS/WEEK)

TLDR is looking for an engineer/researcher at a major AI lab or startup to help write for 1M+ subscribers. Our curators have been invited to Google I/O and OpenAI DevDay, scouted for Tier 1 VCs, and get early access to unreleased TLDR products. Learn more [26].

ANTHROPIC NEARS $1.5 BILLION JOINT VENTURE WITH WALL STREET FIRMS

The investors want to create a company that helps teach businesses how to incorporate AI across their operations.

VLLM ROUTING AND KV

One global vLLM pool is a poor default for mixed traffic.

MORE WAYS TO ADD REFERENCES WHEN USING AI TO GENERATE OR EDIT IMAGES IN FIGMA MAKE

Figma Make introduced new voice-to-text functionality that allows users to dictate prompts directly in chat with automatic text cleaning before submission. The update includes question cards for structured decision-making, version history tracking with instant revert capabilities, and a one-click context reset feature. A new Zapier connector enables integration with Google Drive, Microsoft Office, Zoom, and over 9,000 other applications.

NEXT YEAR'S IPHONE PRO MODELS COULD GET ALL-NEW DESIGN, QUAD-CURVED DISPLAY

Apple's rumored 20th anniversary iPhone in 2027 is now expected to be a major redesign of the iPhone Pro and Pro Max models rather than a separate standalone device. The new design could feature a bezel-free quad-curved display and under-display camera technology, signaling a shift in Apple's biggest innovations toward its mainstream Pro lineup instead of introducing a one-off special edition.

‘THIS IS FINE' CREATOR SAYS AI STARTUP STOLE HIS ART

KC Green, creator of the iconic “This is fine” comic, has accused AI startup Artisan of stealing his artwork for a subway ad promoting its AI sales tool “Ava the AI BDR.” The altered ad changed the comic's caption to “my pipeline is on fire,” prompting Green to publicly denounce the campaign and say he is seeking legal representation. The controversy adds to Artisan's growing reputation for provocative marketing, following earlier “Stop hiring humans” billboards that also sparked backlash.

THE PSYCHOLOGY OF NUDGES: WHY THE SMALLEST DESIGN ELEMENT CAN SHIFT THE BIGGEST OUTCOMES

Design nudges — micro-signals such as defaults, color hierarchies, and progress indicators — silently shape user behavior with an outsized psychological impact. While not inherently unethical, they cross a line when they remove informed consent, mislead users, or favor business interests over user welfare. An auditing framework covering intent, choice clarity, emotional triggers, and cultural sensitivity is proposed to help designers wield nudges with empathy and accountability.

WHO OWNS YOUR DESIGN SYSTEM?

A design system team's structure significantly impacts its success, with three main models: decentralized, centralized, and external. Decentralized models allow multiple teams to contribute directly, providing agility and product-focused solutions, but they risk redundancy and inconsistency. The key is choosing an organizational structure that matches your team's strengths and collaboration style.