TL;DR

Meta CEO Mark Zuckerberg dropped a manifesto detailing his vision for AI and how Meta plans to do things differently. Yes, manifestos are trending and we’re not mad about it.

Some in the industry see AI as a way to automate nearly everything, leaving people to live off a universal income. It’s an idea championed by voices like Elon Musk and OpenAI’s Sam Altman.

But Zuckerberg doesn’t want you to have the heebie-jeebies when it comes to AI, he wants you to feel empowered by all that it can do for you. AI should be something that:

A welcome message in a world obsessed with the AI doomsday narrative.

Meta’s latest earnings sent a clear message: AI isn’t just hype—it’s paying the bills.

The company’s aggressive AI push is already improving its core business. Smarter ad tools powered by AI helped deliver $47.5 billion in Q2 revenue, up 22% year-over-year, crushing expectations.

Investors loved it: shares jumped more than 11% after hours, over $100 billion in market value overnight.

And a huge bill to pay? Zuckerberg’s AI hiring spree. Meta’s CFO said AI hiring will be the second largest driver of next year’s expenses, topped only by AI infrastructure spend.

Recent reports pull back the curtain on this summer’s AI hiring frenzy, which is starting to look like the NBA offseason. Here’s what it really means to be one of the most coveted minds in tech:

Being a nerd has never paid this well, and no one knows where the ceiling is.

As one colleague joked after his business-partner accepted an irresistible Meta offer:

👓 One last neural nugget: Meta believes your AI companion’s true home is in wearable smart glasses, not a screen.

We previously broke down the big bets on AI wearables, including Meta’s Ray-Bans. From breakthrough hits to spectacular fails, read the full rundown.

Excel has ruled business analysis for decades, but it may finally have a real challenger. Meet Shortcut, a new AI-powered tool from MIT spinout Fundamental Research Labs. If you’ve ever wished Excel could do the heavy lifting, Shortcut delivers. It keeps the familiar interface you know while adding agentic AI to handle complex models and workflows.

Here’s what makes Shortcut stand out:

Founder Nico Christie says Shortcut outperforms first-year analysts at McKinsey and Goldman in head-to-head tests. It won 89% of the time, even when the humans had 10 times more time to complete the task. If you work in Excel every day, this tool could quickly become your new secret weapon.

Turn your how-to’s and process flows into beautiful step-by-step video guides automatically with Guidde.

Stop wasting time on repetitive explanations. Guidde’s AI creates stunning video guides in seconds—11x faster.

How it works: Click capture on the browser extension, and Guidde auto-generates step-by-step video guides with visuals, voiceover, and a call to action.

Best part? It’s 100% free.

👉 Download the extension now

Microsoft and Apple crushed quarterly earnings, flexing big-time growth. Microsoft briefly hit a $4T market cap after pulling in $76.4B (up 18%), powered by 39% Azure growth and Copilot’s 100M users. Apple posted $94B (up 10%), thanks to a 13% iPhone boost and strong services, though iPads and wearables slipped. Both are doubling down on AI, with Microsoft pouring $30B+ into data centers and Apple promising deeper AI integration.

Mayo Clinic is teaming up with Nvidia’s SuperPOD to give pancreatic cancer detection a major upgrade. Normally, half of cases aren’t caught until stage four, leaving survival rates at just 13%. But Mayo’s AI, trained on over a million slides, can spot cancer 438 days earlier than current methods, with 97% accuracy compared to humans’ 50%. Still in trials, the tech could redefine early detection when paired with radiologists.

Google is hopping on board the EU’s voluntary AI code of practice, designed to prep developers for the sweeping AI Act. The move comes as Meta bowed out, calling the rules “overreach,” while Anthropic happily signed. The code asks companies to document AI systems, avoid pirated content, and honor content-owner requests for high-risk AI applications. Google still worries the act could stifle innovation but says the updated code is a step up from earlier drafts.

Researchers from Anthropic, UC Berkeley, and Truthful AI have uncovered a surprising phenomenon called “subliminal learning,” where AI models secretly pass down behaviors to other models, even when the training data looks completely neutral.

Here’s how it works:

This hidden transfer showed up across formats like number sequences, code snippets, and math explanations. Even after strict filtering to remove any obvious signals, the behavior persisted.

The good news? The effect only occurred when the teacher and student shared the same architecture (ie both are versions of ChatGPT). Training across LLM models did not result in any subliminal learning.

While preventing misalignment still proves a headache for developers, this discovery is a big step toward understanding the black-box models we all love but barely understand.