AI models typically learn by memorizing patterns, then researchers bolt on reasoning as an afterthought. A new method called Reinforcement Pre-Training flips this approach—teaching models to think during basic training instead.
Good Morning from San Francisco, Meta's AI app turned users into accidental exhibitionists. 😬 People shared medical questions, legal
Meta users think they're chatting privately with AI. Instead, they're broadcasting medical questions, legal troubles, and relationship problems to the world through a public feed that many don't realize exists.
Good Morning from San Francisco,
Meta just dropped $14.8 billion for half of Scale AI 💸. That's Silicon Valley's priciest acqui-hire masquerading as an investment. Zuckerberg got so fed up with his AI teams lagging behind OpenAI that he's throwing cash at a 28-year-old data labeling CEO 🤖.
Meta's Llama 4 flopped. Their "Behemoth" model got delayed. Now they're paying premium prices for glorified human contractors in an industry racing toward automation.
Desperation has a price tag 💰.
Meanwhile, The Browser Company killed Arc to birth Dia—an AI browser that watches everything you do online 👁️.
It reads every tab. Learns your writing style. Accesses your logged-in accounts.
CEO Josh Miller believes AI will replace traditional browsers within five years, so he's betting the company on surveillance-based browsing 🕵️. Arc's clever design couldn't crack mainstream adoption. Now they're testing whether users will trade privacy for convenience ⚖️.
Stay curious,
Marcus Schuler
Meta has agreed to pay $14.8 billion for a 49% stake in Scale AI, in what amounts to Silicon Valley's most expensive acqui-hire disguised as an investment. Mark Zuckerberg's frustration with Meta's AI performance has reached the point where he's willing to throw billions at Scale AI CEO Alexandr Wang to join his company.
The deal comes after Meta's Llama 4 models flopped spectacularly. Zuckerberg has grown frustrated that rivals like OpenAI appear to be further ahead than Meta in underlying AI models and consumer-facing apps. The company even delayed releasing its flagship "Behemoth" model due to performance concerns.
Scale AI built its business on the decidedly unglamorous work of data labeling. Think armies of contractors around the world teaching AI systems the difference between cats and dogs. A big chunk of Meta's $15 billion investment requires the startup to provide future work to Mark Zuckerberg's firm, essentially prepaying for years of data services.
Wang will lead Meta's new "superintelligence" team while maintaining CEO duties at Scale. The 28-year-old has cultivated Pentagon connections and positioned himself as a China hawk, making him an attractive asset for Meta's defense ambitions.
The investment follows a trendy pattern of quasi-acquisitions. Microsoft pulled similar moves with Inflection AI and OpenAI, while Google snatched talent from Character.AI. These deals let companies avoid regulatory scrutiny while hollowing out competitors.
Why this matters:
• Meta just paid premium prices for what amounts to very expensive human labor in an industry racing toward automation
• The deal reveals how desperate established tech companies have become to acquire AI talent, even at absurd valuations
Read on, my dear:
Prompt
A black woman in her 20s sits calmly in a subway train wearing thin cat eye sunglasses, while everyone else around her is blurred in motion, symbolizing stillness in chaos, Realism, moody ambient lighting
The Browser Company released Dia today, an AI browser that monitors everything you do online. The catch? They killed Arc browser development to build it.
Dia sits in your sidebar like a digital assistant that never looks away. It reads every tab you open. It learns your writing style. It accesses sites where you're logged in. After weeks of use, it knows your patterns better than your coworkers do.
This isn't just another browser with AI features bolted on. The company stopped building new features for Arc entirely. CEO Josh Miller believes AI will replace traditional browsers within five years.
Dia's memory feature sounds familiar. Microsoft tried something similar with Recall, which caused a privacy scandal earlier this year. Screenshots of everything users did. Personal data sitting exposed for hackers.
The Browser Company says Dia is different because it's opt-in. But they haven't explained how they'll secure all this personal data. That's concerning when the AI can read your browsing history, access your logged-in accounts, and copy your communication style.
Arc tried to reinvent browsing through clever design. Tabs in the sidebar. Combined bookmarks and browsing. Organizational features everywhere. Tech reviewers loved it. Regular users didn't care.
Miller admits Arc "fell short" of mainstream adoption. Too complex. Too novel. People didn't want to relearn how to browse the web.
So they made a radical bet. Instead of improving Arc, they're betting everything on AI conversation replacing traditional interfaces. Every major browser maker is adding AI features, but few are willing to kill their existing product.
Dia represents a choice about computing's future. Do we want AI assistants that know everything about us? Will convenience beat privacy concerns?
The early signs are mixed. Dia only works on Mac for existing Arc users. Everyone else joins a waitlist. The company isn't sharing usage numbers or user feedback.
Whether this pivot succeeds depends on solving what Arc couldn't: reaching mainstream users while handling their data responsibly. The Browser Company had one innovative browser that failed to find an audience. Now they have an AI assistant that might know too much.
Why this matters:
• The Browser Company's all-in AI pivot shows how artificial intelligence is reshaping entire software categories—browsers might become conversation interfaces rather than tab-based tools.
• Dia's success will test whether users actually want AI assistants with total access to their digital lives, or if privacy concerns limit adoption of the most intrusive features.
Read on, my dear:
Gamma is an AI tool that creates presentations, websites, and documents. You enter a topic, and Gamma automatically turns it into a finished presentation with text, images, and design.
The tool works without design skills. You simply write what you need - for example "Marketing plan for restaurants" - and get a complete presentation back. Gamma selects appropriate colors, fonts, and layouts.
You can use three methods:
The tool has free and paid versions. The free version is sufficient for most users. Gamma does hours of work in just a few minutes.
Sample Task: Type "Marketing strategy for small businesses" and Gamma will create a complete presentation with slides, visuals, and content in under 2 minutes.
Disney and Universal filed a 110-page lawsuit against Midjourney on Wednesday, marking Hollywood's first major legal assault on generative AI companies. The entertainment giants accuse the AI image generator of training on their copyrighted characters without permission, then letting millions of users create knockoff versions of Darth Vader, Elsa, and the Minions.
The lawsuit calls Midjourney "the quintessential copyright free-rider and a bottomless pit of plagiarism." It includes dozens of side-by-side comparisons showing original characters next to AI-generated copies that look remarkably similar. The studios claim they tried negotiating with Midjourney first, but the company refused to cooperate unlike other AI platforms.
Midjourney trains its AI by scraping images from across the internet without asking permission or paying creators. This practice has already triggered lawsuits from authors, artists, and news organizations. But Disney and Universal represent the biggest names in Hollywood to join the legal fight.
The lawsuit targets the platform itself, not individual users who generate images. Disney general counsel Horacio Gutierrez struck a measured tone, saying the company is "bullish on AI technology" but "piracy is piracy" regardless of whether it's done by an AI company.
This case could force AI companies to pay licensing fees for training data instead of scraping copyrighted content for free. The lawsuit tests whether decades-old copyright law can protect creators in the age of artificial intelligence. Other studios may follow Disney's lead, potentially reshaping how the entire AI industry builds its training datasets.
Why this matters:
Read on, my dear:
Meta is suing Joy Timeline HK Limited, the company behind CrushAI, a nudify app that ran over 5,000 ads on Instagram and Facebook despite violating platform policies. The lawsuit comes after 404 Media reported that 90 percent of the app's traffic came from Meta's platforms, where users could upload photos to generate non-consensual nude images using AI.
Researchers found a critical flaw in Microsoft 365 Copilot that let hackers access sensitive company data just by sending an email. The attack, called "EchoLeak," tricks the AI into following hidden instructions and extracting confidential files without users clicking anything or knowing they've been compromised.
SAG-AFTRA suspended its 320-day strike against major video game companies after reaching a tentative deal that includes AI protections for voice and performance capture workers. The agreement delivers over 24% wage increases and requires companies to get consent and provide compensation before using digital replicas of performers' voices or likenesses.
Apple will expand family safety features in iOS 26, letting parents share children's age ranges with apps instead of exact birthdates and requiring approval for new contacts in messaging. The update also extends content filtering and communication safety protections to teens aged 13-17, who previously only got basic restrictions.
Nvidia will build its first industrial AI cloud platform in Germany to help companies like BMW and Mercedes-Benz with design simulation and logistics management. CEO Jensen Huang announced plans for 20 AI factories across Europe and promised to increase the continent's AI computing capacity by ten times within two years.
OpenAI has held talks with Saudi Arabia's sovereign wealth fund and India's Reliance Industries about joining its $40 billion funding round, according to The Information. The ChatGPT maker is seeking hundreds of millions from each investor to fund model development and its ambitious Stargate infrastructure project, with SoftBank leading the round.
Almost a third of Bitcoin's supply sits in centralized treasuries controlled by governments, ETFs, and public companies, according to new research from Gemini and Glassnode. These entities now hold 6.1 million BTC worth $668 billion, marking a 924% increase over the past decade as Bitcoin's price climbed from under $1,000 to over $100,000.
Amsterdam spent years and half a million euros building an AI system to fairly detect welfare fraud, following every best practice in the ethical AI playbook. The Smart Check algorithm still showed bias against different groups during testing, forcing officials to scrap the project and return to human caseworkers who also discriminate—just against different people.
Wikipedia scrapped its AI-generated summary experiment after editors called it a "very bad idea" that would harm the site's reputation. The feature would have shown machine-generated article summaries at the top of pages, but editors argued it undermined Wikipedia's collaborative editing model and risked introducing unreliable content.
Two Polish entrepreneurs just proved you don't need venture capital to build a $15M+ ARR business. NeuronWriter quietly conquered the AI content optimization space while Silicon Valley burned billions chasing unicorn valuations.
The Founders Paweł Sokołowski and Damian Kiełbasa launched in 2019 from Lublin, Poland with zero VC dollars. Operating lean with 2-10 employees, they spotted a gap: manual SERP analysis was eating marketers alive. Sokołowski brought decade-deep SEO expertise; Kiełbasa delivered the technical chops. Pure bootstrap hustle.
The Product Semantic SEO meets AI writing in one brutal combo. Analyzes top 30 Google results, extracts ranking patterns, then generates optimized content using GPT-4o and Claude Sonnet 3.5. Real-time optimization scores, competitor analysis, and WordPress integration. Supports 170+ languages. Does what Surfer SEO does for one-third the price. 💪
The Competition Facing venture-backed giants like Surfer SEO ($69-299/month), Clearscope ($170+), and MarketMuse ($149). NeuronWriter's strategy? Deliver 70-80% functionality at 25% cost. Decathlon, Volkswagen, and Electrolux chose savings over Silicon Valley swagger. AppSumo named it 2023 Tool of the Year.
Financing Zero disclosed VC funding. Bootstrapped through AppSumo lifetime deals and subscription tiers ($25-127/month). 25,000 customers generating estimated $15M+ ARR. No board meetings, no burn rate, pure profit-driven growth.
The Future ⭐⭐⭐⭐ Sustainability beats unicorn dreams. While competitors hemorrhage venture cash, NeuronWriter builds real value. Conservative growth model thrives in current funding winter. Enterprise expansion and agentic AI integration ahead.
Bootstrap beats billions. Sometimes. 🚀
Good Morning from San Francisco, Meta's AI app turned users into accidental exhibitionists. 😬 People shared medical questions, legal
Good Morning from San Francisco, ☀️ OpenAI cut o3 model prices by 80%. Input tokens dropped from $10 to $2 per
Good Morning from San Francisco, Meta burned $10 billion chasing AI talent after losing its entire research team. Zuckerberg spent
Good Morning from San Francisco, Apple researchers just shattered AI's reasoning hype. Those fancy "thinking" models
Get tomorrow's intel today. Join the clever kids' club. Free forever.
