🧬 Two ex-DeepMind startups. $1.3B. One day.

The superintelligence talent war just went nuclear, and Amazon quietly killed one of its biggest bets.

May 13, 2026

In partnership with

My fellow AI explorers

We've got Trump sitting across from Xi in Beijing, talking chips. We've got Meta throwing $100 million packages at researchers like they're candy on Halloween. We've got Amazon killing one of its own chatbots two years after launch. And we've got a 4-month-old startup with barely 20 people raising $650 million to build AI that improves itself without any humans in the loop… Lots to uncover.

In today’s edition:

  • 🤖 Meta's Alexandr Wang defends the $100M talent war

  • 🛒 Amazon kills Rufus and bets everything on Alexa

  • 🔐 WhatsApp gets a private mode for AI chats

  • 🧬 Two ex-DeepMind startups raise $1.3B in a single day

  • 🌏 Trump and Xi talk AI chips in Beijing

The Architecture Behind AI-Native Revenue Automation

In our new white paper, The Architecture Behind AI-Native Revenue Automation, Tabs CTO Deepak Bapat breaks down what it actually takes to apply AI to revenue workflows without breaking the books.

You’ll learn why probabilistic reasoning isn’t enough for finance, how Tabs pairs LLMs with deterministic logic, and why a unified Commercial Graph is the foundation for scalable, audit-ready automation. From contract interpretation to cash application, this paper goes deep on where AI belongs—and where it absolutely doesn’t.

If you’re evaluating AI for billing, collections, or revenue operations, this is the architecture perspective most vendors won’t show you.

AI

Two ex-DeepMind startups just raised a combined $1.3 billion in a single day. No, that is not a typo.

Recursive Superintelligence officially emerged from stealth today after raising $650 million at a $4.65 billion valuation. The round was led by Google's venture arm, GV, and Greycroft, with AMD Ventures and Nvidia also writing checks. Founded by former Salesforce chief scientist Richard Socher and ex-DeepMind researcher Tim Rocktäschel, alongside former OpenAI researchers, the company has fewer than 30 employees and zero product.

What they do have is an audacious thesis: AI that recursively improves itself, without humans in the loop, automating the entire AI development pipeline end-to-end.

The company put it plainly in their launch post: "The fastest path to superintelligence will be realized by AI that recursively improves itself."

Here's what that actually means:

  • Recursive self-improvement means the AI designs its own training runs, selects its own data, and sets its own research direction

  • The goal is to automate what today requires armies of PhDs at labs like OpenAI and Anthropic

  • They're calling this the "third and perhaps final stage of neural networks"

Meanwhile, on the same day, Ineffable Intelligence, which is founded by David Silver (the man behind AlphaGo), announced a new engineering-level partnership with Nvidia to build AI systems that learn purely by trial and error. Silver is betting that reinforcement learning, not language models, is the real path to superhuman intelligence. Nvidia CEO Jensen Huang personally endorsed the collaboration, calling superlearners "the next frontier of AI."

Two companies. Founded months ago. Combined raise: $1.3 billion. Combined employees: under 60.

🔮 Prediction: The frontier AI race is no longer a two-horse contest between OpenAI and Anthropic. A wave of ex-DeepMind and ex-OpenAI researchers is building competing architectures: RL/AI-first, self-improving, and radically different from today's LLM paradigm. The next breakthrough might not come from the biggest lab. It might come from the leanest one, with new startups looking to polish their own AI-first, recursively improving model.

Meta

The AI talent war is getting surreal, and Meta's right at the center of it.

Meta AI chief Alexandr Wang, the 28-year-old former CEO of Scale AI who joined Meta as part of a $14.3 billion deal, pushed back this week on the narrative that his team is purely money-motivated. Reports surfaced that Meta was dangling compensation packages north of $100 million to lure researchers from rivals like OpenAI, Google, and xAI.

Wang's response? "I think it's an incorrect assumption to think that the researchers are just money motivated or anything."

Here's what's actually happening:

  • Meta has been offering some researchers packages reaching $100M+, including $200M to Apple's former AI models team lead, Ruoming Pang

  • Wang was himself recruited as part of Meta's $14.3 billion investment in Scale AI, making him one of the most expensive executive hires in tech history

  • Meta's pitch to talent isn't just cash. It's compute. Unlimited GPU access at a time when chips are in short supply everywhere else

But not everyone is buying it.

Several top OpenAI researchers have reportedly turned down the offers, questioning whether their work would actually have an impact inside Zuckerberg's machine.

What's undeniable: the market for AI talent has become completely detached from any normal economic logic. Sam Altman reportedly called out Meta for offering bonuses as high as $100 million to his own employees. And with VCs pouring $18.8 billion into AI startups in 2026 alone, researchers now have three options: stay at a Big Lab, go to a well-funded startup, or start their own.

💬 The real tension: Inside major labs, the pressure to ship commercial products is crowding out exploratory research. That's exactly why David Silver and Tim Rocktäschel left DeepMind in the first place. The talent exodus isn't just about money. It's about freedom. And right now, the startup world is selling freedom at a premium.

AI Privacy

Meta quietly dropped something important today that deserves a lot more attention than it's getting.

WhatsApp launched Incognito Chat with Meta AI, which is a mode where your conversation with the AI chatbot is fully private. Not "private" in the Instagram DM sense. Actually private. Messages are processed in a secure environment that even Meta itself cannot read. Nothing is saved. Everything disappears when you close the session.

Here's what's actually new:

  • Messages are processed in a secure environment that Meta cannot access, built on top of WhatsApp's Private Processing technology

  • Conversations are not saved by default and disappear when you exit the session

  • Other apps have tried incognito-style modes, but they could still see the questions and answers. This one genuinely can't

  • Coming next: "Side Chat." It’s a private AI mode that runs alongside any WhatsApp conversation without disrupting it

Will Cathcart, Meta's head of WhatsApp, framed it well: "We're starting to ask a lot of meaningful questions about our lives with AI systems, and it doesn't always feel like you should have to share the information behind those questions with the companies that run those AI systems."

Think about what people actually ask AI: health symptoms they're embarrassed about, financial situations they haven't told their partners, questions about mental health, relationship advice, career fears. All of it has been going to servers that companies log and, in many cases, use for training.

🔮 Prediction: This is a Trojan horse for mainstream AI adoption. People who refused to use AI chatbots because of privacy fears now have a legitimate safe space. Expect Meta AI's usage numbers to jump significantly in the countries where WhatsApp dominates, which is most of the world outside the U.S. The privacy-first AI UX isn't a niche feature. It's the next competitive moat.

30-Second AI Play

🛒 How to Use the New Alexa for Shopping Before Your Competitors Do

Amazon just quietly handed you a research and deal-finding tool that most people don't know exists yet. Here's how to get ahead of it today:

  1. Open the Amazon app or Amazon.com. Look for the new cursive "A" icon that replaced the old Rufus button.

  2. Ask it like you'd ask a friend. Try: "What's a good skincare routine for men under $50?" or "When did I last order AA batteries?" It pulls from your full order history.

  3. Use the comparison mode. Ask it to compare two specific products side by side before you buy.

  4. Set a price alert. Tell Alexa for Shopping to notify you when a specific item drops to your target price. It'll ping you across app, mobile, and Echo Show.

  5. Build a shopping guide. Ask it to create a custom buying guide for a specific project (e.g., "I'm setting up a home gym, what do I need?") and it'll generate a curated list.

🔍 Why it matters: Amazon says Rufus had 300 million users in 2025, and this is meaningfully better. Unlike ChatGPT or Gemini's shopping tools, Alexa for Shopping has access to your full order history, real-time stock levels, and actual delivery estimates. Other AI shopping agents are guessing. This one knows you.

Other Relevant AI News!

🌏 Trump and Xi are sitting across from each other in Beijing right now, and AI chips are on the agenda. Trump announced he's approving Nvidia's H200 chips for sale to certain Chinese customers, with 25% of sales revenue going to the U.S. government. It’s the first major reversal of Biden-era export restrictions and a signal that the AI chip war is entering the negotiation phase.

🩺 650,000 American doctors are now quietly using an AI tool called OpenEvidence to make clinical decisions, write patient discharge notes, and prep for licensing exams, and two-thirds of U.S. physicians are active users, according to NBC News reporting. The most widely adopted AI tool in medicine that nobody outside medicine is talking about.

🛒 Amazon officially discontinued the Rufus chatbot just two years after launch, replacing it with Alexa for Shopping: a unified AI shopping agent that combines Rufus's recommendation engine with Alexa+'s memory across all devices. No Prime membership required. Available now in the U.S.

💸 Recursive Superintelligence co-founder Tim Rocktäschel cited Stanisław Lem's concept of the "information barrier," or the point where knowledge grows so fast humans can no longer keep up, as the founding thesis of his company. And the company just raised $650M to break through it with AI that rewrites its own algorithms endlessly.

Golden Nuggets

  • 🧬 The most dangerous question in AI right now isn't "can it think?" It's "can it improve itself?" And two newly funded labs just bet $1.3B that the answer is yes

  • 💰 Meta is paying $100M-plus per researcher while a 20-person startup raises $650M with no product. We are deep in the irrational exuberance phase of superintelligence and it is genuinely fascinating to watch

  • 🔐 WhatsApp's incognito AI mode isn't a gimmick. It's the privacy unlock that will bring the next billion users to AI chatbots

Would love to hear your thoughts on ChatGPT Atlas! Send me your thoughts by replying to this email (yes, I read them all :)

Until our next AI rendezvous,

Anthony | Founder of Uncover AI