An AI Too Scary To Release. We Bought 100 chicks.

6min Read | Watch | Listen

Hey there Builders!

Last night I'm at the dinner table with my Mutual Aid group. We do this weekly dinner swap where we fully cook for a family one night, they fully cook for us another night.

Two meals we cook. Two meals get cooked for us. That leaves three dinners a week we handle ourselves.

Community connection. Money saved. Less stress around "what's for dinner" 7 nights a week.

So I made a shepherd's pie with lamb we raised and harvested ourselves. My daughter just turned 8, and we're all joking around at the table about fast food.

I crack a joke about McDonald's.

My daughter looks at me and says, "Dad... what's McDonald's?"

My neighbor looks over at me, dead serious, and says, "Good job, parents. Kid's 8 years old and doesn't even know what McDonald's is."

Why You Should Give All the Cares:

• The AI That's Too Dangerous to Release: Anthropic built a model called Mythos that finds security holes no human can. They're not letting anyone use it yet.

• 3:30 AM Builds and 20 Million Views: What I've been waking up before dawn to create, and why I'm packaging it up for you.

• 100 Meat Chicks and a Chicken Tractor: Grocery prices pushed us to the edge. We're filling our freezers ourselves this spring.

The AI Model That Has Its Own Creators Scared

Anthropic, the company behind Claude, has been sitting on something big since February.

They built a model called Mythos. 10 trillion parameters. The largest AI model ever created. And they made a deliberate choice: don't release it to the public.

Instead, they assembled a coalition called Project Glasswing, bringing together Amazon, Apple, Google, Microsoft, Nvidia, JP Morgan, Crowdstrike, and others. The mission: let these companies harden their software before Mythos gets into wider hands.

Because Mythos already found thousands of zero-day vulnerabilities (security flaws nobody knew existed) in every major operating system and web browser on the planet.

Some of these bugs had been hiding for decades. One vulnerability in OpenBSD, considered one of the most secure systems in the world, had been sitting there for 27 years.

It found and chained together could give an attacker full control of a machine. It found a 16-year-old bug in FFmpeg, the library that runs video across most of the internet.

The patches Mythos wrote were so clean that FFmpeg's maintainers thought they came from human developers.

Here's where it gets unsettling. During testing, Mythos broke out of sandboxed environments it wasn't supposed to escape. One Anthropic researcher got an email from an instance of Mythos while eating lunch at the park (yes, really). That instance wasn't supposed to have internet access.

The Anthropic team's own words to describe this model: "frightening" and "spooky."

And the wild part: it's also the most well-aligned model they've ever built. Harder to manipulate, harder to trick, harder to prompt-inject than anything else out there.

We can't use it yet. But it exists. And the people who built it are losing sleep over what it means.

Matthew Berman did a solid deep dive on this if you want the full breakdown: Mythos Is Real and It Scares Me

What I'm Building at 3:30 AM

I've been getting up before dawn multiple days this week. Not because I had to. Because I couldn't stop.

I'm packaging up a Viral Content Kit based on the tools and systems I used to hit 20 million views in Feb while testing everything out.

It includes an app, a playbook, and a set of AI-powered skills that plug into Claude's Cowork mode.

I'm rolling it out in tiers so people can grab what fits:

• If you just want your posts to look right, there's a simple app for that.

• If you already use AI but the output feels generic, the playbook gives you the frameworks and patterns (the stuff that actually moves the needle).

• If you want the full automated system built on Claude, that's available too.

• And if you want me to just set it all up for you, we can do that.

I'm setting it up for a couple clients right now, sanding down the rough edges, and I hope to start sharing it more widely next week.

If you’re interested reply back to this email and ill get you on the waitlist.

100 Meat Chicks and a Freezer Full of Real Food

Our mutual aid group meets tonight, and Saturday we're building a chicken tractor.

We're about to buy 100 meat chicks and raise them to fill our freezers.

If you've been to the grocery store lately, you already know why. the prices are out of control, and this is something we've been working toward for a few months now.

It's go time. I'll be taking video and sharing the process as we build it out.

Tech Corner

• HeyGen Video Avatars: HeyGen just dropped their latest video avatar tech. All it needs is 15 seconds of footage of someone and it can mimic them on video almost perfectly. Video generation across the board is getting close to flawless. AI video slop is going to be a thing of the past within a year.

• Meta Launches Muse Spark: Meta is back in the AI race. Muse Spark is the first major model from their new Meta Superintelligence Labs. it powers WhatsApp, Instagram, Facebook, Messenger, and Meta's AI glasses soon.

• CIA Used AI Tech to Rescue a Downed Airman: The CIA reportedly used a classified AI tool called Ghost Murmur to help locate and rescue a downed U.S. airman in Iran. That actually happened.

The Chatbot to Coworker replay is now available at a discount, but it goes back to full price next week.

If you want to understand how to actually use AI so it works for you (not the other way around), this is your window. Grab it here.

Today’s Takeaways:

1. Feed Your People, Know Your People: A dinner swap costs nothing to start and builds the kind of community that money can't buy.

2. AI Just Lapped Us: Mythos exists, it's better than humans at finding software vulnerabilities, and we can't even use it yet. Pay attention.

3. Build Your Own Supply Chain: Grocery prices aren't going back down. Raising your own food is becoming less of a hobby and more of a strategy.

I love reading the responses to these newsletters.

Hit reply and tell me what landed this week.

What are you thinking about AI right now? Are you growing your own food? Are you building something at weird hours because you can't help yourself?

I want to hear it.

Stay sharp and stay free,

Joshua | The Holistic Tech Wizard

What did you think about today’s newsletter?

Reply with some feedback or just a 1- 5 ⭐

P.S. If this landed for you, pass it on to someone who needs it. One forward can change someone's week.